NS-3-TUTORIAL(1)
✨ NAME ✨
okay bestie, welcome to ns-3-tutorial(1) — aka the ns-3 Tutorial, and it's literally the blueprint for getting into network simulation. 💖 Primary documentation for the ns-3 project comes in five iconic forms:
- ns-3 Doxygen: Documentation of the public APIs of the simulator
- Tutorial (this document), Manual, and Model Library for the latest release and development tree
- ns-3 wiki
This document is written in reStructuredText for Sphinx and is maintained in the doc/tutorial directory of ns-3’s source code. 🎀
🚀 QUICK START 🚀
This section is optional, for readers who want to get ns-3 up and running as quickly as possible. Readers may skip forward to the Introduction chapter, followed by the Getting Started chapter, for a lengthier coverage of all of this material. No cap, it's just the speedrun route if you're impatient like me. 😌
💫 Brief Summary 💫
ns-3 is a discrete-event simulator typically run from the command line. It is written directly in C++, not in a high-level modeling language; simulation events are simply C++ function calls, organized by a scheduler. Periodt.
An ns-3 user will obtain the ns-3 source code (see below), compile it into shared (or static) libraries, and link the libraries to main() programs that he or she authors. The main() program is where the specific simulation scenario configuration is performed and where the simulator is run and stopped. Several example programs are provided, which can be modified or copied to create new simulation scenarios. Users also often edit the ns-3 library code (and rebuild the libraries) to change its behavior.
ns-3 has optional Python bindings for authoring scenario configuration programs in Python (and using a Python-based workflow); this quick start does not cover those aspects. iykyk, we're keeping it C++ for now.
🧚 Prerequisites 🧚
ns-3 has various optional extensions, but the main features just require a C++ compiler (g++ or clang++) and Python (version 3.6 or above); the Python is needed for the build system. We focus in this chapter only on getting ns-3 up and running on a system supported by a recent C++ compiler and Python runtime support.
For Linux, use either g++ or clang++ compilers. For macOS, use clang++ (available in Xcode or Xcode Command Line Tools). For Windows, we recommend to either use a Linux virtual machine, or the Windows Subsystem for Linux. choose your fighter wisely 💅
📥 Downloading ns-3 📥
ns-3 is distributed in source code only (some binary packages exist but they are not maintained by the open source project). There are two main ways to obtain the source code: 1) downloading the latest release as a source code archive from the main ns-3 web site, or 2) cloning the Git repository from GitLab.com. These two options are described next; either one or the other download option (but not both) should be followed. pick one bestie, don't do both, that's messy.
🍓 Downloading the Latest Release 🍓
- Download the latest release from https://www.nsnam.org/releases/latest
- Unpack it in a working directory of your choice.
$ tar xjf ns-allinone-3.35.tar.bz2- Change into the ns-3 directory directly; e.g.
$ cd ns-allinone-3.35/ns-3.35The ns-allinone directory has some additional components but we are skipping over them here; one can work directly from the ns-3 source code directory. we're keeping it focused, bestie.
🌈 Cloning ns-3 from GitLab.com 🌈
You can perform a Git clone in the usual way:
$ git clone https://gitlab.com/nsnam/ns-3-dev.gitIf you are content to work with the tip of the development tree; you need only to cd into ns-3-dev; the master branch is checked out by default.
$ cd ns-3-devIf instead you want to try the most recent release (version 3.35 as of this writing), you can checkout a branch corresponding to that git tag:
$ git checkout -b ns-3.35-branch ns-3.35slay, now you're on the right branch ✨
🔥 Building and testing ns-3 🔥
Once you have obtained the source either by downloading a release or by cloning a Git repository, the next step is to configure the build using the Waf build system that comes with ns-3. There are several options to control the build, but enabling the example programs and the tests, for a default debug build profile (with debugging symbols and support for ns-3 logging) is what is usually done at first:
$ ./waf configure --enable-examples --enable-testsThen, use Waf to build ns-3:
$ ./waf buildOnce complete, you can run the unit tests to check your build:
$ ./test.pyAll tests should either PASS or be SKIPped. At this point, you have a working ns-3 simulator. From here, you can start to run programs (look in the examples directory). To run the first tutorial program, whose source code is located at examples/tutorial/first.cc, use Waf to run it (by doing so, the ns-3 shared libraries are found automatically):
$ ./waf --run firstTo view possible command-line options, specify the –PrintHelp argument:
$ ./waf --run 'first --PrintHelp'To continue reading about the conceptual model and architecture of ns-3, the tutorial chapter Conceptual Overview would be the next natural place to skip to, or you can learn more about the project and the various build options by continuing directly with the Introduction and Getting Started chapters.
and that's the vibe check for part one, bestie — more coming, so don't go anywhere 💖
💖 INTRODUCTION 💖
Okay bestie, so the ns-3 simulator is a discrete-event network simulator and it's targeted primarily for research and educational use. The ns-3 project? Started in 2006, it's an open-source project developing ns-3. Iconic behavior honestly. ✨
The whole point of this tutorial is to introduce new ns-3 users to the system in a structured way. Because ngl, it is sometimes genuinely difficult for new users to glean essential information from detailed manuals and to convert this information into working simulations. So in this tutorial, we will build several example simulations, introducing and explaining key concepts and features as we go. We're doing this together, hand in hand. 🫶
As the tutorial unfolds, we will introduce the full ns-3 documentation and provide pointers to source code for those interested in delving deeper into the workings of the system. And we also provide a quick start guide for those who are comfortable diving right in without too much documentation. Main character energy. 💅
A few key points are worth noting at the onset:
- ns-3 is open-source, and the project strives to maintain an open environment for researchers to contribute and share their software. 🌈
- ns-3 is not a backwards-compatible extension of ns-2; it is a new simulator. The two simulators are both written in C++ but ns-3 is a new simulator that does not support the ns-2 APIs. Not a glow-up, a whole new girl. 💀
🦋 About ns-3 🦋
ns-3 has been developed to provide an open, extensible network simulation platform, for networking research and education. In brief, ns-3 provides models of how packet data networks work and perform, and provides a simulation engine for users to conduct simulation experiments.
Some of the reasons to use ns-3 include:
- to perform studies that are more difficult or not possible to perform with real systems
- to study system behavior in a highly controlled, reproducible environment
- to learn about how networks work 🌸
Users will note that the available model set in ns-3 focuses on modeling how Internet protocols and networks work, but ns-3 is not limited to Internet systems; several users are using ns-3 to model non-Internet-based systems. Versatile queen. 👑
Many simulation tools exist for network simulation studies. Below are a few distinguishing features of ns-3 in contrast to other tools.
- ns-3 is designed as a set of libraries that can be combined together and also with other external software libraries. While some simulation platforms provide users with a single, integrated graphical user interface environment in which all tasks are carried out, ns-3 is more modular in this regard. Several external animators and data analysis and visualization tools can be used with ns-3. However, users should expect to work at the command line and with C++ and/or Python software development tools. Bestie, we're terminal girls now. 🎀
- ns-3 is primarily used on Linux or macOS systems, although support exists for BSD systems and also for Windows frameworks that can build Linux code, such as Windows Subsystem for Linux, or Cygwin. Native Windows Visual Studio is not presently supported although a developer is working on future support. Windows users may also use a Linux virtual machine. 🙈
- ns-3 is not an officially supported software product of any company. Support for ns-3 is done on a best-effort basis on the ns-3-users forum (
ns-3-users@googlegroups.com). 💖
💜 For ns-2 Users 💜
For those familiar with ns-2 (a popular tool that preceded ns-3), the most visible outward change when moving to ns-3 is the choice of scripting language.
Programs in ns-2 are scripted in OTcl and results of simulations can be visualized using the Network Animator nam. It is not possible to run a simulation in ns-2 purely from C++ (i.e., as a main() program without any OTcl). Moreover, some components of ns-2 are written in C++ and others in OTcl.
In ns-3, the simulator is written entirely in C++, with optional Python bindings. Simulation scripts can therefore be written in C++ or in Python. New animators and visualizers are available and under current development. Since ns-3 generates pcap packet trace files, other utilities can be used to analyze traces as well. In this tutorial, we will first concentrate on scripting directly in C++ and interpreting results via trace files. 🦋
But there are similarities as well (both, for example, are based on C++ objects, and some code from ns-2 has already been ported to ns-3). We will try to highlight differences between ns-2 and ns-3 as we proceed in this tutorial.
A question that we often hear is "Should I still use ns-2 or move to ns-3?" In this author's opinion, unless the user is somehow vested in ns-2 (either based on existing personal comfort with and knowledge of ns-2, or based on a specific simulation model that is only available in ns-2), a user will be more productive with ns-3 for the following reasons:
- ns-3 is actively maintained with an active, responsive users mailing list, while ns-2 is only lightly maintained and has not seen significant development in its main code tree for over a decade. 🔥
- ns-3 provides features not available in ns-2, such as a implementation code execution environment (allowing users to run real implementation code in the simulator)
- ns-3 provides a lower base level of abstraction compared with ns-2, allowing it to align better with how real systems are put together. Some limitations found in ns-2 (such as supporting multiple types of interfaces on nodes correctly) have been remedied in ns-3. Slay.
ns-2 has a more diverse set of contributed modules than does ns-3, owing to its long history. However, ns-3 has more detailed models in several popular areas of research (including sophisticated LTE and WiFi models), and its support of implementation code admits a very wide spectrum of high-fidelity models. Users may be surprised to learn that the whole Linux networking stack can be encapsulated in an ns-3 node, using the Direct Code Execution (DCE) framework. ns-2 models can sometimes be ported to ns-3, particularly if they have been implemented in C++. ✨
If in doubt, a good guideline would be to look at both simulators (as well as other simulators), and in particular the models available for your research, but keep in mind that your experience may be better in using the tool that is being actively developed and maintained (ns-3). 💅
🌈 Contributing 🌈
ns-3 is a research and educational simulator, by and for the research community. It will rely on the ongoing contributions of the community to develop new models, debug or maintain existing ones, and share results. There are a few policies that we hope will encourage people to contribute to ns-3 like they have for ns-2:
- Open source licensing based on GNU GPLv2 compatibility
- wiki
- Contributed Code page, similar to ns-2's popular Contributed Code page
- Use of GitLab.com including issue tracker <‐
https://www.gitlab.com/nsnam
We realize that if you are reading this document, contributing back to the project is probably not your foremost concern at this point, but we want you to be aware that contributing is in the spirit of the project and that even the act of dropping us a note about your early experience with ns-3 (e.g. "this tutorial section was not clear…"), reports of stale documentation or comments in the code, etc. are much appreciated. 🫶
The preferred way to submit patches is either to fork our project on GitLab.com and generate a Merge Request, or to open an issue on our issue tracker and append a patch. Periodt. 👑
🎀 Tutorial Organization 🎀
The tutorial assumes that new users might initially follow a path such as the following:
- Try to download and build a copy;
- Try to run a few sample programs;
- Look at simulation output, and try to adjust it.
As a result, we have tried to organize the tutorial along the above broad sequences of events. iykyk. 💖
🌐 RESOURCES 🌐
🌸 The Web 🌸
okay bestie, there are like a few essential resources that literally every ns-3 user has to know about, no cap. the main web site lives at https://www.nsnam.org and it's your home base for basic info about the whole ns-3 system. want the detailed documentation? that's at https://www.nsnam.org/documentation/ through the main site too. you can also find documents about the system architecture right there on that same page. ✨
there's a Wiki that complements the main ns-3 web site and you'll find it at https://www.nsnam.org/wiki/. that's where the user and developer FAQs live, plus troubleshooting guides, third-party contributed code, papers, etc. it's giving treasure chest fr fr 💖
the source code itself can be found and browsed at GitLab.com: https://gitlab.com/nsnam/. over there you'll find the current development tree in the repository named ns-3-dev. and if you're feeling nostalgic (or just curious), past releases and experimental repositories from the core developers can also be found at the project's old Mercurial site at http://code.nsnam.org. 🦋
🎀 Git 🎀
so here's the thing — complex software systems need some way to manage the organization and changes to the underlying code and documentation. there are many ways to do this feat, and you've probably heard of some of the systems people use for it. until recently, the ns-3 project used Mercurial as its source code management system, but in December 2018, it switched to using Git. 💅
now, you do not need to know much about Git to complete this tutorial, but we still recommend becoming familiar with Git and using it to access the source code. GitLab.com provides resources to get you started at: https://docs.gitlab.com/ee/gitlab-basics/. 👑
🔥 Waf 🔥
once you've got source code downloaded to your local system, you're gonna need to compile that source to produce usable programs. and just like with source code management, there are many tools available to do this. probably the most well known of these tools is make. and along with being the most well known, make is probably the most difficult to use in a very large and highly configurable system. because of that, lots of alternatives have been developed — and recently these systems have been developed using the Python language. 😭
the build system Waf is what's used on the ns-3 project. it's one of the new generation of Python-based build systems. and the bestie-est part? you will not need to understand any Python to build the existing ns-3 system. periodt. 🫶
for those interested in the gory details of Waf, the Waf book is available at https://waf.io/book/ and the current code at https://gitlab.com/ita1024/waf/. 💀
🧚 Development Environment 🧚
as mentioned above, scripting in ns-3 is done in C++ or Python. most of the ns-3 API is available in Python, but the models are written in C++ in either case. a working knowledge of C++ and object-oriented concepts is assumed in this document. we'll take some time to review some of the more advanced concepts or possibly unfamiliar language features, idioms and design patterns as they appear. but we don't want this tutorial to devolve into a C++ tutorial, so we do expect a basic command of the language. there are a wide number of sources of information on C++ available on the web or in print. 🌈
if you're new to C++, you may want to find a tutorial- or cookbook-based book or web site and work through at least the basic features of the language before proceeding. for instance, this tutorial. 🙈
on Linux, the ns-3 system uses several components of the GNU "toolchain" for development. a software toolchain is the set of programming tools available in the given environment. for a quick review of what's included in the GNU toolchain see, http://en.wikipedia.org/wiki/GNU_toolchain. ns-3 uses gcc, GNU binutils, and gdb. however, we do not use the GNU build system tools, neither make nor autotools. we use Waf for these functions. ⭐
on macOS, the toolchain used is Xcode. ns-3 users on a Mac are strongly encouraged to install Xcode and the command-line tools packages from the Apple App Store, and to look at the ns-3 installation wiki for more information (https://www.nsnam.org/wiki/Installation). 🍓
typically an ns-3 author will work in Linux or a Unix-like environment. for those running under Windows, there do exist environments which simulate the Linux environment to various degrees. the ns-3 project has in the past (but not presently) supported development in the Cygwin environment for these users. see http://www.cygwin.com/ for details on downloading, and visit the ns-3 wiki for more information about Cygwin and ns-3. MinGW is presently not officially supported. another alternative to Cygwin is to install a virtual machine environment such as VMware server and install a Linux virtual machine. 💜
🦋 Socket Programming 🦋
we're gonna assume a basic facility with the Berkeley Sockets API in the examples used in this tutorial. if you're new to sockets, we recommend reviewing the API and some common usage cases. for a good overview of programming TCP/IP sockets we recommend TCP/IP Sockets in C, Donahoo and Calvert. 📚
there's an associated web site that includes source for the examples in the book, which you can find at: http://cs.baylor.edu/~donahoo/practical/CSockets/.
if you understand the first four chapters of the book (or for those who don't have access to a copy of the book, the echo clients and servers shown in the website above) you'll be in good shape to understand the tutorial. there's a similar book on Multicast Sockets, Multicast Sockets, Makofske and Almeroth, that covers material you may need to understand if you look at the multicast examples in the distribution. ✨
🚀 GETTING STARTED 🚀
this section is aimed at getting a user to a working state starting with a machine that may never have had ns-3 installed. it covers supported platforms, prerequisites, ways to obtain ns-3, ways to build ns-3, and ways to verify your build and run simple programs. main character energy, let's gooo. 💅
👑 Overview 👑
ns-3 is built as a system of software libraries that work together. user programs can be written that links with (or imports from) these libraries. user programs are written in either the C++ or Python programming languages. 🔥
ns-3 is distributed as source code, meaning that the target system needs to have a software development environment to build the libraries first, then build the user program. ns-3 could in principle be distributed as pre-built libraries for selected systems, and in the future it may be distributed that way, but at present, many users actually do their work by editing ns-3 itself, so having the source code around to rebuild the libraries is useful. if someone would like to undertake the job of making pre-built libraries and packages for operating systems, please contact the ns-developers mailing list. 🫶
in the following, we'll look at three ways of downloading and building ns-3. the first is to download and build an official release from the main web site. the second is to fetch and build development copies of a basic ns-3 installation. the third is to use an additional build tool to download more extensions for ns-3. we'll walk through each since the tools involved are slightly different. 🎀
experienced Linux users may wonder at this point why ns-3 is not provided like most other libraries using a package management tool? although there exist some binary packages for various Linux distributions (e.g. Debian), most users end up editing and having to rebuild the ns-3 libraries themselves, so having the source code available is more convenient. we will therefore focus on a source installation in this tutorial. 😭
for most uses of ns-3, root permissions are not needed, and the use of a non-privileged user account is recommended. 🌸
⭐ Prerequisites ⭐
the entire set of available ns-3 libraries has a number of dependencies on third-party libraries, but most of ns-3 can be built and used with support for a few common (often installed by default) components: a C++ compiler, an installation of Python, a source code editor (such as vim, emacs, or Eclipse) and, if using the development repositories, an installation of Git source code control system. most beginning users need not concern themselves if their configuration reports some missing optional features of ns-3, but for those wishing a full installation, the project provides a wiki that includes pages with many useful hints and tips. one such page is the "Installation" page, with install instructions for various systems, available at https://www.nsnam.org/wiki/Installation. 🌈
the "Prerequisites" section of this wiki page explains which packages are required to support common ns-3 options, and also provides the commands used to install them for common Linux or macOS variants. 🧚
you may want to take this opportunity to explore the ns-3 wiki a bit, or the main web site at https://www.nsnam.org, since there is a wealth of information there. 💖
as of the most recent ns-3 release (ns-3.35), the following tools are needed to get started with ns-3:
- Prerequisite: C++ compiler — Package/version:
clang++org++(g++version 7 or greater) - Prerequisite: Python — Package/version:
python3version >=3.6 - Prerequisite: Git — Package/version: any recent version (to access ns-3 from GitLab.com)
- Prerequisite: tar — Package/version: any recent version (to unpack an ns-3 release)
- Prerequisite: bunzip2 — Package/version: any recent version (to uncompress an ns-3 release)
to check the default version of Python, type python -V. to check the default version of g++, type g++ -v. if your installation is missing or too old, please consult the ns-3 installation wiki for guidance. 💀
okay so from this point forward we're just gonna assume you're on Linux, macOS, or some Linux-emulation situation, and that you've got all the prerequisites from above sorted. 💅
Note: the ns-3 build system (Waf, which we'll meet in a sec) does NOT tolerate spaces in the installation path. Like, at all. So make sure you're downloading into a directory where nothing in the full path has a space in it. 🙈
For example, do not use a path like this one, because one of the parent directories has a space in its name:
$ pwd
/home/user/5G simulations/ns-3-allinone/ns-3-dev✨ Downloading a release of ns-3 as a source archive ✨
This one's for the new user who just wants to download and mess around with the most recently released and packaged version of ns-3. The ns-3 project publishes its releases as compressed source archives, which people sometimes call a tarball. A tarball is just a specific format of software archive where a bunch of files get bundled together and then usually compressed. Downloading ns-3 via tarball is so simple: pick a release, download it, uncompress it. that's the whole vibe. 🎀
Let's say you, as a user, want to build ns-3 in a local directory called workspace. If you go with the workspace-directory approach, you can grab a copy of a release by typing this into your Linux shell (swap in the right version numbers, obviously):
$ cd
$ mkdir workspace
$ cd workspace
$ wget https://www.nsnam.org/release/ns-allinone-3.35.tar.bz2
$ tar xjf ns-allinone-3.35.tar.bz2See that wget up there? It's a command-line tool for fetching stuff off the web; if you don't have it installed, you can just use a browser for this step. 🦋
Once you've done all that, if you change into the ns-allinone-3.35 directory you should see a bunch of files and directories:
$ cd ns-allinone-3.35
$ ls
bake constants.py ns-3.35 README
build.py netanim-3.108 pybindgen-0.22.0 util.pyAnd now you're ready to build the base ns-3 distribution, so you can skip ahead to the section on building ns-3. ✨
💖 Downloading ns-3 using Git 💖
The ns-3 code lives in Git repositories on the GitLab.com service at https://gitlab.com/nsnam/. The group name nsnam is what organizes all the various repositories the open source project uses. 👑
The simplest way to get going with Git repositories is to fork or clone the ns-3-allinone environment. That's a set of scripts that handles downloading and building the most commonly used subsystems of ns-3 for you. If you're new to Git, the words fork and clone might be giving absolute nonsense to you — and if so, bestie, we recommend you just clone (aka make your own replica) of the repository over on GitLab.com, like this:
$ cd
$ mkdir workspace
$ cd workspace
$ git clone https://gitlab.com/nsnam/ns-3-allinone.git
$ cd ns-3-allinoneAt this point your view of the ns-3-allinone directory is slightly different from what we described above with a release archive; it should look a little something like this:
$ ls
build.py constants.py download.py README util.pyNotice the download.py script sitting right there — that one's gonna go fetch the ns-3 and related sourcecode for you. 🌸 At this point you've got a choice: either download the most recent development snapshot of ns-3:
$ python3 download.pyor specify a release of ns-3, using the -n flag to name a release number:
$ python3 download.py -n ns-3.35After this step, the additional repositories of ns-3, bake, pybindgen, and netanim will be downloaded into the ns-3-allinone directory. 💫
🧚 Downloading ns-3 Using Bake 🧚
Those two techniques above (source archive, or the ns-3-allinone repository via Git) are great for getting the most basic installation of ns-3 with a couple of addons — pybindgen for generating Python bindings, and netanim for network animiations. The third repository provided by default in ns-3-allinone is called bake. 🍓
Bake is a tool for coordinated software building from multiple repositories, and it was developed for the ns-3 project. You can use Bake to fetch development versions of the ns-3 software, and to download and build extensions to the base ns-3 distribution — stuff like the Direct Code Execution environment, Network Simulation Cradle, the ability to create new Python bindings, and various ns-3 "apps". So if you're envisioning that your ns-3 installation might use advanced or optional features, this is lowkey the installation path for you. 👑
In recent ns-3 releases, Bake has been included in the release tarball. The configuration file that ships in the released version will let you download any software that was current at the time of the release. So like, the version of Bake distributed with the ns-3.30 release can be used to fetch components for that ns-3 release or earlier, but it can't be used to fetch components for later releases (unless the bakeconf.xml package description file gets updated).
You can also grab the most recent copy of bake by typing this into your Linux shell (assuming you've installed Git):
$ cd
$ mkdir workspace
$ cd workspace
$ git clone https://gitlab.com/nsnam/bake.gitAs the git command runs, you should see something like this pop up:
Cloning into 'bake'...
remote: Enumerating objects: 2086, done.
remote: Counting objects: 100% (2086/2086), done.
remote: Compressing objects: 100% (649/649), done.
remote: Total 2086 (delta 1404), reused 2078 (delta 1399)
Receiving objects: 100% (2086/2086), 2.68 MiB | 3.82 MiB/s, done.
Resolving deltas: 100% (1404/1404), done.Once the clone command finishes, you should have a directory called bake, and its contents should look a little something like this:
$ cd bake
$ ls
bake bakeconf.xml bake.py doc examples generate-binary.py test TODONotice that you've downloaded some Python scripts, a Python module called bake, and an XML configuration file. 💖 Next step is gonna be using those scripts to download and build the ns-3 distribution of your choice.
There are a few configuration targets available:
- ns-3.35: the module corresponding to the release; it will download components similar to the release tarball.
- ns-3-dev: a similar module but using the development code tree.
- ns-allinone-3.35: the module that includes other optional features such as bake build system, netanim animator, and pybindgen.
- ns-3-allinone: similar to the released version of the allinone module, but for development code.
The current development snapshot (unreleased) of ns-3 may be found at https://gitlab.com/nsnam/ns-3-dev.git. The developers do try to keep these repositories in consistent, working states, but it is a development area with unreleased code hanging out in it, so you might wanna consider staying with an official release if you don't need any newly-introduced features. 🫶
You can find the latest version of the code either by inspecting the repository list, or by heading to the "ns-3 Releases" web page and clicking the latest release link. We'll proceed in this tutorial example with ns-3.35. ⭐
Now we're gonna use the bake tool to pull down the various pieces of ns-3 you'll be using. But first, a word about running bake. 🎀
Bake works by downloading source packages into a source directory, and installing libraries into a build directory. You can run bake by referencing the binary, but if you choose to run bake from outside the directory it was downloaded into, it's advisable to put bake into your path, like this (Linux bash shell example). First, change into the bake directory, then set the following environment variables:
$ export BAKE_HOME=`pwd`
$ export PATH=$PATH:$BAKE_HOME:$BAKE_HOME/build/bin
$ export PYTHONPATH=$PYTHONPATH:$BAKE_HOME:$BAKE_HOME/build/libThis puts the bake.py program into the shell's path, and lets other programs find executables and libraries created by bake. Although several bake use cases don't require setting PATH and PYTHONPATH like this, full builds of ns-3-allinone (with the optional packages) typically do. 💅
Step into the workspace directory and type this into your shell:
$ ./bake.py configure -e ns-3.35Next, we'll ask bake to check whether we've got enough tools to download the various components. Type:
$ ./bake.py checkYou should see something like this:
> Python - OK
> GNU C++ compiler - OK
> Mercurial - OK
> Git - OK
> Tar tool - OK
> Unzip tool - OK
> Make - OK
> cMake - OK
> patch tool - OK
> Path searched for tools: /usr/local/sbin /usr/local/bin /usr/sbin /usr/bin /sbin /bin ...In particular, download tools like Git and Mercurial are our principal concerns at this point, since they're what let us fetch the code. Please install any missing tools at this stage, in the usual way for your system (if you're able to), or contact your system administrator as needed to get those tools installed. You can also (and yeah, the manpage literally just leaves that sentence hanging there — not me, that's the source fr fr)
Next, try to download the software:
$ ./bake.py downloadwhich should yield something like:
>> Searching for system dependency setuptools - OK
>> Searching for system dependency libgoocanvas2 - OK
>> Searching for system dependency gi-cairo - OK
>> Searching for system dependency pygobject - OK
>> Searching for system dependency pygraphviz - OK
>> Searching for system dependency python-dev - OK
>> Searching for system dependency qt - OK
>> Searching for system dependency g++ - OK
>> Searching for system dependency cxxfilt - OK
>> Searching for system dependency setuptools - OK
>> Searching for system dependency gi-cairo - OK
>> Searching for system dependency gir-bindings - OK
>> Searching for system dependency pygobject - OK
>> Searching for system dependency cmake - OK
>> Downloading netanim-3.108 - OK
>> Downloading pybindgen-0.22.0 (target directory:pybindgen) - OK
>> Downloading ns-3.35 (target directory:ns-3.35) - OKAll of that means three sources have been downloaded. 🔥 Check the source directory now and type ls; you should see:
$ cd source
$ ls
netanim-3.108 ns-3.35 pybindgenAnd now you're ready to build the ns-3 distribution. ✨
✨ Building ns-3 ✨
okay bestie, just like with downloading, there's more than one way to build ns-3 🌸 and here's the one thing we really need you to internalize: ns-3 gets built by a build tool called Waf, described below. Most users will end up working most directly with Waf, but there are also some convenience scripts to get you started or to orchestrate more complicated builds. So please have a look at build.py and building with bake before you read about Waf below. 💅
🛠️ Building with build.py 🛠️
Note: this build step is only available from a source archive release described above; not from downloading via git or bake. 🫶
When you're working from a released tarball, a convenience script that's available as part of ns-3-allinone can orchestrate a simple build of components. This program is called build.py ✨. It'll get the project configured for you in the most commonly useful way. But heads up: more advanced configuration and work with ns-3 will typically involve using the native ns-3 build system, Waf, which we'll introduce later in this tutorial.
If you downloaded using a tarball you should have a directory called something like ns-allinone-3.35 under your ~/workspace directory. Type the following:
$ ./build.py --enable-examples --enable-testsBecause we're working with examples and tests in this tutorial, and because they are not built by default in ns-3, the arguments for build.py tell it to build them for us. The program also defaults to building all available modules. Later, you can build ns-3 without examples and tests, or eliminate the modules that aren't necessary for your work, if you wish. 🦋
You'll see lots of compiler output messages displayed as the build script builds the various pieces you downloaded. First, the script will attempt to build the netanim animator, then the pybindgen bindings generator, and finally ns-3. Eventually you should see the following:
Waf: Leaving directory '/path/to/workspace/ns-allinone-3.35/ns-3.35/build'
'build' finished successfully (6m25.032s)
Modules built:
antenna aodv applications
bridge buildings config-store
core csma csma-layout
dsdv dsr energy
fd-net-device flow-monitor internet
internet-apps lr-wpan lte
mesh mobility mpi
netanim (no Python) network nix-vector-routing
olsr point-to-point point-to-point-layout
propagation sixlowpan spectrum
stats tap-bridge test (no Python)
topology-read traffic-control uan
virtual-net-device visualizer wave
wifi wimax
Modules not built (see ns-3 tutorial for explanation):
brite click openflow
Leaving directory ./ns-3.35Now, about that portion about modules not built:
Modules not built (see ns-3 tutorial for explanation):
brite clickThis just means that some ns-3 modules that have dependencies on outside libraries may not have been built, or that the configuration specifically asked not to build them. It does not mean that the simulator didn't build successfully, or that it will provide wrong results for the modules listed as being built. 😌
🧁 Building with bake 🧁
If you used bake above to fetch source code from project repositories, you may continue to use it to build ns-3. Type:
$ ./bake.py buildand you should see something like:
>> Building pybindgen-0.22.0 - OK
>> Building netanim-3.108 - OK
>> Building ns-3.35 - OKThere may be failures to build all components, but the build will proceed anyway if the component is optional. For example, a recent portability issue has been that castxml may not build via the bake build tool on all platforms; in this case, the line will show something like:
>> Building castxml - Problem
> Problem: Optional dependency, module "castxml" failed
This may reduce the functionality of the final build.
However, bake will continue since "castxml" is not an essential dependency.
For more information call bake with -v or -vvv, for full verbose mode.However, castxml is only needed if one wants to generate updated Python bindings, and most users do not need to do so (or to do so until they're more involved with ns-3 changes), so such warnings might be safely ignored for now. 🙈
If there happens to be a failure, please have a look at what the following command tells you; it may give a hint as to a missing dependency:
$ ./bake.py showThis will list out the various dependencies of the packages you're trying to build. 💜
💅 Building with Waf 💅
Up to this point, we've used either the build.py script, or the bake tool, to get started with building ns-3. These tools are useful for building ns-3 and supporting libraries, and they call into the ns-3 directory to call the Waf build tool to do the actual building. An installation of Waf is bundled with the ns-3 source code. Most users quickly transition to using Waf directly to configure and build ns-3. So, to proceed, please change your working directory to the ns-3 directory that you have initially built. 🎀
It's not strictly required at this point, but it will be valuable to take a slight detour and look at how to make changes to the configuration of the project. Probably the most useful configuration change you can make will be to build the optimized version of the code. By default you have configured your project to build the debug version. Let's tell the project to make an optimized build. To explain to Waf that it should do optimized builds that include the examples and tests, you will need to execute the following commands:
$ ./waf clean
$ ./waf configure --build-profile=optimized --enable-examples --enable-testsThis runs Waf out of the local directory (which is provided as a convenience for you). The first command to clean out the previous build is not typically strictly necessary but is good practice (but see Build Profiles, below); it will remove the previously built libraries and object files found in directory build/. When the project is reconfigured and the build system checks for various dependencies, you should see output that looks similar to the following:
Setting top to : /home/ns3user/workspace/bake/source/ns-3-dev
Setting out to : /home/ns3user/workspace/bake/source/ns-3-dev/build
Checking for 'gcc' (C compiler) : /usr/bin/gcc
Checking for cc version : 7.3.0
Checking for 'g++' (C++ compiler) : /usr/bin/g++
Checking for compilation flag -march=native support : ok
Checking for compilation flag -Wl,--soname=foo support : ok
Checking for compilation flag -std=c++11 support : ok
Checking boost includes : headers not found, please provide a --boost-includes argument (see help)
Checking boost includes : headers not found, please provide a --boost-includes argument (see help)
Checking for program 'python' : /usr/bin/python
Checking for python version >= 2.3 : 2.7.15
python-config : /usr/bin/python-config
Asking python-config for pyembed '--cflags --libs --ldflags' flags : yes
Testing pyembed configuration : yes
Asking python-config for pyext '--cflags --libs --ldflags' flags : yes
Testing pyext configuration : yes
Checking for compilation flag -fvisibility=hidden support : ok
Checking for compilation flag -Wno-array-bounds support : ok
Checking for pybindgen location : ../pybindgen (guessed)
Checking for python module 'pybindgen' : 0.21.0
Checking for pybindgen version : 0.21.0
Checking for code snippet : yes
Checking for types uint64_t and unsigned long equivalence : no
Checking for code snippet : no
Checking for types uint64_t and unsigned long long equivalence : yes
Checking for the apidefs that can be used for Python bindings : gcc-LP64
Checking for internal GCC cxxabi : complete
Checking for python module 'pygccxml' : not found
Checking for click location : not found
Checking for program 'pkg-config' : /usr/bin/pkg-config
Checking for 'gtk+-3.0' : not found
Checking for 'libxml-2.0' : yes
checking for uint128_t : not found
checking for __uint128_t : yes
Checking high precision implementation : 128-bit integer (default)
Checking for header stdint.h : yes
Checking for header inttypes.h : yes
Checking for header sys/inttypes.h : not found
Checking for header sys/types.h : yes
Checking for header sys/stat.h : yes
Checking for header dirent.h : yes
Checking for header stdlib.h : yes
Checking for header signal.h : yes
Checking for header pthread.h : yes
Checking for header stdint.h : yes
Checking for header inttypes.h : yes
Checking for header sys/inttypes.h : not found
Checking for library rt : yes
Checking for header sys/ioctl.h : yes
Checking for header net/if.h : yes
Checking for header net/ethernet.h : yes
Checking for header linux/if_tun.h : yes
Checking for header netpacket/packet.h : yes
Checking for NSC location : not found
Checking for 'sqlite3' : not found
Checking for header linux/if_tun.h : yes
Checking for python module 'gi' : 3.26.1
Checking for python module 'gi.repository.GObject' : ok
Checking for python module 'cairo' : ok
Checking for python module 'pygraphviz' : 1.4rc1
Checking for python module 'gi.repository.Gtk' : ok
Checking for python module 'gi.repository.Gdk' : ok
Checking for python module 'gi.repository.Pango' : ok
Checking for python module 'gi.repository.GooCanvas' : ok
Checking for program 'sudo' : /usr/bin/sudo
Checking for program 'valgrind' : not found
Checking for 'gsl' : not found
python-config : not found
Checking for compilation flag -fstrict-aliasing support : ok
Checking for compilation flag -fstrict-aliasing support : ok
Checking for compilation flag -Wstrict-aliasing support : ok
Checking for compilation flag -Wstrict-aliasing support : ok
Checking for program 'doxygen' : /usr/bin/doxygen
---- Summary of optional NS-3 features:
Build profile : optimized
Build directory :
BRITE Integration : not enabled (BRITE not enabled (see option --with-brite))
DES Metrics event collection : not enabled (defaults to disabled)
Emulation FdNetDevice : enabled
Examples : enabled
File descriptor NetDevice : enabled
GNU Scientific Library (GSL) : not enabled (GSL not found)
Gcrypt library : not enabled (libgcrypt not found: you can use libgcrypt-config to find its location.)
GtkConfigStore : not enabled (library 'gtk+-3.0 >= 3.0' not found)
MPI Support : not enabled (option --enable-mpi not selected)
NS-3 Click Integration : not enabled (nsclick not enabled (see option --with-nsclick))
NS-3 OpenFlow Integration : not enabled (Required boost libraries not found)
Network Simulation Cradle : not enabled (NSC not found (see option --with-nsc))
PlanetLab FdNetDevice : not enabled (PlanetLab operating system not detected (see option --force-planetlab))
PyViz visualizer : enabled
Python API Scanning Support : not enabled (Missing 'pygccxml' Python module)
Python Bindings : enabled
Real Time Simulator : enabled
SQlite stats data output : not enabled (library 'sqlite3' not found)
Tap Bridge : enabled
Tap FdNetDevice : enabled
Tests : enabled
Threading Primitives : enabled
Use sudo to set suid bit : not enabled (option --enable-sudo not selected)
XmlIo : enabled
'configure' finished successfully (6.387s)Okay bestie, eyes back on that last chunk of the output. 👀 Some ns-3 options are not enabled by default, and some straight up need support from the underlying system before they'll work. Like, to enable XmlTo, the library libxml-2.0 has to actually be found on your system. If that library isn't there, the matching ns-3 feature just doesn't get enabled, and it tells you so with a message. ✨
Also peep this: there's a feature that uses the program sudo to set the suid bit on certain programs. That one is not enabled by default, so it's reported as “not enabled.” And when you want to reprint that whole summary of which optional features are enabled? Just use the --check-config option to waf. 💅
Now go ahead and switch back to the debug build that includes the examples and tests. 🎀
$ ./waf clean
$ ./waf configure --build-profile=debug --enable-examples --enable-testsThe build system is now configured, and you can build the debug versions of the ns-3 programs by simply typing: 🌸
$ ./wafSo yes, the steps above made you build the ns-3 part of the system twice — but now you know how to change the configuration and build optimized code. That's the whole point, iykyk. 😌
There's also a command for checking which profile is currently active for an already configured project:
$ ./waf --check-profile
Waf: Entering directory \`/path/to/ns-allinone-3.35/ns-3.35/build\'
Build profile: debugThe build.py script we talked about earlier also supports the --enable-examples and enable-tests arguments, but ngl, in general it does not directly support other waf options. For example, this will not work: 💀
$ ./build.py --disable-pythonand it will result in:
build.py: error: no such option: --disable-pythonBut bestie, the special operator -- is here to save the day — you can use it to pass additional options through to waf. So instead of the above, this works: 🫶
$ ./build.py -- --disable-pythonbecause it generates the underlying command ./waf configure --disable-python. Periodt. ✨
Here are a few more introductory tips about Waf.
🛠 Handling build errors 🛠
ns-3 releases get tested against the most recent C++ compilers available in the mainstream Linux and macOS distributions at the time of the release. But here's the thing — over time, newer distributions drop, with newer compilers, and those newer compilers tend to be way more pedantic about warnings. 😭 And ns-3 configures its build to treat all warnings as errors. So it's sometimes the case that, if you're running an older release version on a newer system, a compiler warning will just make the build fail. Not the vibe.
For instance, ns-3.28 was released prior to Fedora 28, which shipped a new major version of gcc (gcc-8). Building ns-3.28 or older releases on Fedora 28, when Gtk2+ is installed, will result in an error like this:
/usr/include/gtk-2.0/gtk/gtkfilechooserbutton.h:59:8: error: unnecessary parentheses in declaration of ‘__gtk_reserved1’ [-Werror=parentheses]
void (*__gtk_reserved1);In releases starting with ns-3.28.1, there's an option available in Waf to work around these issues. The option disables the inclusion of the ‘-Werror’ flag to g++ and clang++. The option is ‘–disable-werror’ and it must be used at configure time; e.g.:
./waf configure --disable-werror --enable-examples --enable-tests✨ Configure vs. Build ✨
Okay bestie, so some Waf commands are literally only meaningful during the configure phase, and some are valid in the build phase. Not all commands are welcome at all times, it's giving "you can't sit with us" energy. 💅
For example, if you wanted to use the emulation features of ns-3, you might want to enable setting the suid bit using sudo as described above. This turns out to be a configuration-time command, and so you could reconfigure using the following command that also includes the examples and tests:
$ ./waf configure --enable-sudo --enable-examples --enable-testsIf you do this, Waf will have run sudo to change the socket creator programs of the emulation code to run as root. 👑
There are many other configure- and build-time options available in Waf. To explore these options, type:
$ ./waf --helpWe’ll use some of the testing-related commands in the next section.
🎀 Build Profiles 🎀
We already saw how you can configure Waf for debug or optimized builds:
$ ./waf --build-profile=debugThere is also an intermediate build profile, release. -d is a synonym for --build-profile. Lowkey iconic shorthand. ✨
The build profile controls the use of logging, assertions, and compiler optimization:
Build profiles
- Feature → Build Profile
- debug
- Enabled Features:
NS3_BUILD_PROFILE_DEBUG,NS_LOG...,NS_ASSERT... - Code Wrapper Macro:
NS_BUILD_DEBUG(code) - Compiler Flags:
-O0 -ggdb -g3
- Enabled Features:
- release
- Enabled Features:
NS3_BUILD_PROFILE_RELEASE - Code Wrapper Macro:
NS_BUILD_RELEASE(code) - Compiler Flags:
-O3 -g0 -fomit-frame-pointer
- Enabled Features:
- optimized
- Enabled Features:
NS3_BUILD_PROFILE_OPTIMIZED - Code Wrapper Macro:
NS_BUILD_OPTIMIZED(code) - Compiler Flags:
-O3 -g -fstrict-overflow -march=native
- Enabled Features:
- debug
As you can see, logging and assertions are only configured by default in debug builds, although they can be selectively enabled in other build profiles by using the--enable-logsand--enable-assertsflags during Waf configuration time. Recommended practice is to develop your scenario in debug mode, then conduct repetitive runs (for statistics or changing parameters) in optimized build profile. If you have code that should only run in specific build profiles, use the indicated Code Wrapper macro:NS_BUILD_DEBUG (std::cout << "Part of an output line..." << std::flush; timer.Start ()); DoLongInvolvedComputation (); NS_BUILD_DEBUG (timer.Stop (); std::cout << "Done: " << timer << std::endl;)By default Waf puts the build artifacts in thebuilddirectory. You can specify a different output directory with the--outoption, e.g.$ ./waf configure --out=my-build-dirCombining this with build profiles lets you switch between the different compile options in a clean way:$ ./waf configure --build-profile=debug --out=build/debug $ ./waf build ... $ ./waf configure --build-profile=optimized --out=build/optimized $ ./waf build ...This allows you to work with multiple builds rather than always overwriting the last build. When you switch, Waf will only compile what it has to, instead of recompiling everything. When you do switch build profiles like this, you have to be careful to give the same configuration parameters each time. It may be convenient to define some environment variables to help you avoid mistakes:$ export NS3CONFIG="--enable-examples --enable-tests" $ export NS3DEBUG="--build-profile=debug --out=build/debug" $ export NS3OPT=="--build-profile=optimized --out=build/optimized" $ ./waf configure $NS3CONFIG $NS3DEBUG $ ./waf build ... $ ./waf configure $NS3CONFIG $NS3OPT $ ./waf build
🛠️ Compilers and Flags 🛠️
In the examples above, Waf uses the GCC C++ compiler, g++, for building ns-3. However, it’s possible to change the C++ compiler used by Waf by defining the CXX environment variable. For example, to use the Clang C++ compiler, clang++,
$ CXX="clang++" ./waf configure
$ ./waf buildOne can also set up Waf to do distributed compilation with distcc in a similar way:
$ CXX="distcc g++" ./waf configure
$ ./waf buildMore info on distcc and distributed compilation can be found on its project page under the Documentation section. 🦋
To add compiler flags, use the CXXFLAGS_EXTRA environment variable when you configure ns-3.
📦 Install 📦
Waf may be used to install libraries in various places on the system. The default location where libraries and executables are built is in the build directory, and because Waf knows the location of these libraries and executables, it is not necessary to install the libraries elsewhere. It's giving "everything is already cute where it is" energy. ✨
If users choose to install things outside of the build directory, users may issue the ./waf install command. By default, the prefix for installation is /usr/local, so ./waf install will install programs into /usr/local/bin, libraries into /usr/local/lib, and headers into /usr/local/include. Superuser privileges are typically needed to install to the default prefix, so the typical command would be sudo ./waf install. When running programs with Waf, Waf will first prefer to use shared libraries in the build directory, then will look for libraries in the library path configured in the local environment. So when installing libraries to the system, it is good practice to check that the intended libraries are being used. 👑
Users may choose to install to a different prefix by passing the --prefix option at configure time, such as:
./waf configure --prefix=/opt/localIf later after the build the user issues the ./waf install command, the prefix /opt/local will be used.
The ./waf clean command should be used prior to reconfiguring the project if Waf will be used to install things at a different prefix.
In summary, bestie, it is not necessary to call ./waf install to use ns-3. 💖 Most users will never need this command, since Waf will just pick up the current libraries from the build directory — but some users might find it useful if their whole use case involves working with programs outside of the ns-3 directory.
✨ One Waf ✨
There is only one Waf script, and it lives at the top level of the ns-3 source tree. Periodt. As you work, you may find yourself spending a <em>lot</em> of time in scratch/, or deep in src/..., and needing to invoke Waf anyway. You could just remember where you are and invoke Waf like this:
$ ../../../waf ...but that gets tedious, and error prone, and honestly there are better solutions. 🙈
One common way when you're using a text-based editor such as emacs or vim is to open two terminal sessions and use one to build ns-3 and one to edit source code. Main character multitasking energy.
If you only have the tarball, an environment variable can help:
$ export NS3DIR="$PWD"
$ function waff { cd $NS3DIR && ./waf $* ; }
$ cd scratch
$ waff buildIt might be tempting in a module directory to add a trivial waf script along the lines of exec ../../waf. Please don't. 😭 It's confusing to newcomers, and when done poorly it leads to subtle build errors. The solutions above are the way to go. Slay responsibly.
🧪 Testing ns-3 🧪
You can run the unit tests of the ns-3 distribution by running the ./test.py script:
$ ./test.pyThese tests are run in parallel by Waf. You should eventually see a report saying that
92 of 92 tests passed (92 passed, 0 failed, 0 crashed, 0 valgrind errors)This is the important message to check for; failures, crashes, or valgrind errors indicate problems with the code or incompatibilities between the tools and the code. 💀
You will also see the summary output from Waf and the test runner executing each test, which will actually look something like:
Waf: Entering directory `/path/to/workspace/ns-3-allinone/ns-3-dev/build'
Waf: Leaving directory `/path/to/workspace/ns-3-allinone/ns-3-dev/build'
'build' finished successfully (1.799s)
Modules built:
aodv applications bridge
click config-store core
csma csma-layout dsdv
emu energy flow-monitor
internet lte mesh
mobility mpi netanim
network nix-vector-routing ns3tcp
ns3wifi olsr openflow
point-to-point point-to-point-layout propagation
spectrum stats tap-bridge
template test tools
topology-read uan virtual-net-device
visualizer wifi wimax
PASS: TestSuite wifi-interference
PASS: TestSuite histogram
...
PASS: TestSuite object
PASS: TestSuite random-number-generators
92 of 92 tests passed (92 passed, 0 failed, 0 crashed, 0 valgrind errors)This command is typically run by users to quickly verify that an ns-3 distribution has built correctly. (Note that the order of the PASS: ... lines can vary, which is okay. What's important is that the summary line at the end report that all tests passed; none failed or crashed.) 🫶
Both Waf and test.py will split up the job on the available CPU cores of the machine, in parallel. Parallelism? Iconic.
🏃 Running a Script 🏃
We typically run scripts under the control of Waf. This allows the build system to make sure that the shared library paths are set correctly and that the libraries are available at run time. To run a program, simply use the --run option in Waf. Let's run the ns-3 equivalent of the ubiquitous hello world program by typing the following:
$ ./waf --run hello-simulatorWaf first checks to make sure that the program is built correctly and executes a build if required. Waf then executes the program, which produces the following output.
Hello SimulatorCongratulations! You are now an ns-3 user! 🎉👑
What do I do if I don't see the output?
If you see Waf messages indicating that the build was completed successfully, but do not see the "Hello Simulator" output, chances are that you have switched your build mode to optimized in the Building with Waf section, but have missed the change back to debug mode. All of the console output used in this tutorial uses a special ns-3 logging component that is useful for printing user messages to the console. Output from this component is automatically disabled when you compile optimized code – it is "optimized out." If you don't see the "Hello Simulator" output, type the following:
$ ./waf configure --build-profile=debug --enable-examples --enable-teststo tell Waf to build the debug versions of the ns-3 programs that includes the examples and tests. You must still build the actual debug version of the code by typing
$ ./wafNow, if you run the hello-simulator program, you should see the expected output. 💅
🎀 Program Arguments 🎀
To feed command line arguments to an ns-3 program, use this pattern:
$ ./waf --run <ns3-program> --command-template="%s <args>"Substitute your program name for <em><ns3-program></em>, and the arguments for <em><args></em>. The --command-template argument to Waf is basically a recipe for constructing the actual command line Waf should use to execute the program. Waf checks that the build is complete, sets the shared library paths, then invokes the executable using the provided command line template, inserting the program name for the %s placeholder. Chef's kiss. 🤌
If you find the above to be syntactically complicated, a simpler variant exists, which is to include the ns-3 program and its arguments enclosed by single quotes, such as:
$ ./waf --run '<ns3-program> --arg1=value1 --arg2=value2 ...'Another particularly useful example is to run a test suite by itself. Let's assume that a mytest test suite exists (it doesn't). Above, we used the ./test.py script to run a whole slew of tests in parallel, by repeatedly invoking the real testing program, test-runner. To invoke test-runner directly for a single test:
$ ./waf --run test-runner --command-template="%s --suite=mytest --verbose"This passes the arguments to the test-runner program. Since mytest does not exist, an error message will be generated. 😭 To print the available test-runner options:
$ ./waf --run test-runner --command-template="%s --help"🐛 Debugging 🐛
To run ns-3 programs under the control of another utility, such as a debugger (e.g. gdb) or memory checker (e.g. valgrind), you use a similar --command-template="..." form.
For example, to run your ns-3 program hello-simulator with the arguments <em><args></em> under the gdb debugger:
$ ./waf --run=hello-simulator --command-template="gdb %s --args <args>"Notice that the ns-3 program name goes with the --run argument, and the control utility (here gdb) is the first token in the --command-template argument. The --args tells gdb that the remainder of the command line belongs to the "inferior" program. (Some gdb's don't understand the --args feature. In this case, omit the program arguments from the --command-template, and use the gdb command set args.) 🧚
We can combine this recipe and the previous one to run a test under the debugger:
$ ./waf --run test-runner --command-template="gdb %s --args --suite=mytest --verbose"📁 Working Directory 📁
Waf needs to run from its location at the top of the ns-3 tree. This becomes the working directory where output files will be written. But what if you want to keep those files out of the ns-3 source tree? Use the --cwd argument:
$ ./waf --cwd=...It may be more convenient to start with your working directory where you want the output files, in which case a little indirection can help:
$ function waff {
CWD="$PWD"
cd $NS3DIR >/dev/null
./waf --cwd="$CWD" $*
cd - >/dev/null
}This embellishment of the previous version saves the current working directory, cd's to the Waf directory, then instructs Waf to change the working directory back to the saved current working directory before running the program. ✨
We mention this --cwd command for completeness; most users will simply run Waf from the top-level directory and generate the output data files there.
⚡ Running without Building ⚡
As of the ns-3.30 release, a new Waf option was introduced to allow the running of programs while skipping the build step. This can reduce the time to run programs when, for example, running the same program repeatedly through a shell script, or when demonstrating program execution. This option, --run-no-build, behaves the same as the -run option, except that the program and ns-3 libraries will not be rebuilt. No cap.
$ ./waf --run-no-build '<ns3-program> --arg1=value1 --arg2=value2 ...'🏷️ Build version 🏷️
As of the ns-3.32 release, a new Waf configure option --enable-build-version was introduced which inspects the local ns3 git repository during builds and adds version metadata to the core module.
This configuration option has the following prerequisites:
- The ns-3 directory must be part of a local git repository
- The local git repository must have at least one ns-3 release tag
or
- A file named
version.cache, containing version information, is located in thesrc/coredirectory
If these prerequisites are not met, the configuration will fail. 💀
When these prerequisites are met and ns-3 is configured with the --enable-build-version option, the waf command --check-version can be used to query the local git repository and display the current version metadata.
$ ./waf --check-versionWaf will collect information about the build and print out something similar to the output below.
ns-3.33+249@g80e0dd0-dirty-debugIf --check-version is run when -enable-build-version was not configured, an error message indicating that the option is disabled will be displayed instead.
Build version support is not enabled, reconfigure with --enable-build-version flagThe build information is generated by examining the current state of the git repository. The output of --check-version will change whenever the state of the active branch changes. 🦋
The output of --check-version has the following format:
<version_tag>[+closest_tag][+distance_from_tag]@<commit_hash>[-tree_state]-<profile>version_tag— okay soversion_tagcontains the version of the ns-3 code. The version tag is defined as a git tag with the formatns-3*. If multiple git tags match the format, the tag on the active branch which is closest to the current commit is chosen. periodt. 💅closest_tag—closest_tagis similar toversion_tagexcept it is the first tag found, regardless of format. The closest tag is not included in the output whenclosest_tagandversion_taghave the same value. 🫶distance_from_tag—distance_from_tagcontains the number of commits between the current commit andclosest_tag.distance_from_tagis not included in the output when the value is0(i.e. whenclosest_tagpoints to the current commit). 😭commit_hash—commit_hashis the hash of the commit at the tip of the active branch. The value is‘g’followed by the first 7 characters of the commit hash. The‘g’prefix is used to indicate that this is a git hash. iykyk. ⭐tree_state—tree_stateindicates the state of the working tree. When the working tree has uncommitted changes this field has the value‘dirty’. The tree state is not included in the version output when the working tree is clean (e.g. when there are no uncommitted changes). 💀profile— the build profile specified in the--build-profileoption passed towaf configure. 🌸
A new class, named Version, has been added to the core module. The Version class contains functions to retrieve individual fields of the build version as well as functions to print the full build version like --check-version. The build-version-example application provides an example of how to use the Version class to retrieve the various build version fields. See the documentation for the Version class for specifics on the output of the Version class functions. 👑
The version information stored in the Version class is updated every time the git repository changes. This may lead to frequent recompilations/linking of the core module when the --enable-build-version option is configured. heads up, bestie!
build-version-example:
Program Version (according to CommandLine): ns-3.33+249@g80e0dd0-dirty-debug
Version fields:
LongVersion: ns-3.33+249@g80e0dd0-dirty-debug
ShortVersion: ns-3.33+*
BuildSummary: ns-3.33+*
VersionTag: ns-3.33
Major: 3
Minor: 33
Patch: 0
ReleaseCandidate:
ClosestAncestorTag: ns-3.33
TagDistance: 249
CommitHash: g80e0dd0
BuildProfile: debug
WorkingTree: dirtyThe CommandLine class has also been updated to support the --version option which will print the full build version and exit. 🚪✨
./waf --run-no-build "command-line-example --version"
Waf: Entering directory `/g/g14/mdb/gitlab/mdb/ns-3-dev/build/debug'
ns-3.33+249@g80e0dd0-dirty-debugIf the --enable-build-version option was not configured, --version will print out a message similar to --check-version indicating that the build version option is not enabled. no cap.
Source version 💾
An alternative to storing build version information in the ns-3 libraries is to track the source code version used to build the code. When using Git, the following recipe can be added to Bash shell scripts to create a version.txt file with Git revision information, appended with a patch of any changes to that revision if the repository is dirty. The resulting text file can then be saved with any corresponding ns-3 simulation results. 🎀
echo `git describe` > version.txt
gitDiff=`git diff`
if [[ $gitDiff ]]
then
echo "$gitDiff" >> version.txt
fiCONCEPTUAL OVERVIEW 🧠✨
The first thing we need to do before actually starting to look at or write ns-3 code is to explain a few core concepts and abstractions in the system. Much of this may appear transparently obvious to some, but we recommend taking the time to read through this section just to ensure you are starting on a firm foundation. bestie, trust. 🌈
Key Abstractions 🔑
In this section, we’ll review some terms that are commonly used in networking, but have a specific meaning in ns-3. 👀
Node 🖥️
In Internet jargon, a computing device that connects to a network is called a host or sometimes an end system. Because ns-3 is a network simulator, not specifically an Internet simulator, we intentionally do not use the term host since it is closely associated with the Internet and its protocols. Instead, we use a more generic term also used by other simulators that originates in Graph Theory — the node. 🕸️
In ns-3 the basic computing device abstraction is called the node. This abstraction is represented in C++ by the class Node. The Node class provides methods for managing the representations of computing devices in simulations. 💻
You should think of a Node as a computer to which you will add functionality. One adds things like applications, protocol stacks and peripheral cards with their associated drivers to enable the computer to do useful work. We use the same basic model in ns-3. main character energy. 👑
Application 📱
Typically, computer software is divided into two broad classes. System Software organizes various computer resources such as memory, processor cycles, disk, network, etc., according to some computing model. System software usually does not use those resources to complete tasks that directly benefit a user. A user would typically run an application that acquires and uses the resources controlled by the system software to accomplish some goal. 🧠
Often, the line of separation between system and application software is made at the privilege level change that happens in operating system traps. In ns-3 there is no real concept of operating system and especially no concept of privilege levels or system calls. We do, however, have the idea of an application. Just as software applications run on computers to perform tasks in the “real world,” ns-3 applications run on ns-3 Nodes to drive simulations in the simulated world. 🌍➡️🌐
In ns-3 the basic abstraction for a user program that generates some activity to be simulated is the application. This abstraction is represented in C++ by the class Application. The Application class provides methods for managing the representations of our version of user-level applications in simulations. Developers are expected to specialize the Application class in the object-oriented programming sense to create new applications. In this tutorial, we will use specializations of class Application called UdpEchoClientApplication and UdpEchoServerApplication. As you might expect, these applications compose a client/server application set used to generate and echo simulated network packets. 💌
Channel 📡
In the real world, one can connect a computer to a network. Often the media over which data flows in these networks are called channels. When you connect your Ethernet cable to the plug in the wall, you are connecting your computer to an Ethernet communication channel. In the simulated world of ns-3, one connects a Node to an object representing a communication channel. Here the basic communication subnetwork abstraction is called the channel and is represented in C++ by the class Channel. 🔌
The Channel class provides methods for managing communication subnetwork objects and connecting nodes to them. Channels may also be specialized by developers in the object oriented programming sense. A Channel specialization may model something as simple as a wire. The specialized Channel can also model things as complicated as a large Ethernet switch, or three-dimensional space full of obstructions in the case of wireless networks. 🌐
We will use specialized versions of the Channel called CsmaChannel, PointToPointChannel and WifiChannel in this tutorial. The CsmaChannel, for example, models a version of a communication subnetwork that implements a carrier sense multiple access communication medium. This gives us Ethernet-like functionality. ✨
Net Device 🧩
It used to be the case that if you wanted to connect a computer to a network, you had to buy a specific kind of network cable and a hardware device called (in PC terminology) a peripheral card that needed to be installed in your computer. If the peripheral card implemented some networking function, they were called Network Interface Cards, or NICs. Today most computers come with the network interface hardware built in and users don’t see these building blocks. 🧱
A NIC will not work without a software driver to control the hardware. In Unix (or Linux), a piece of peripheral hardware is classified as a device. Devices are controlled using device drivers, and network devices (NICs) are controlled using network device drivers collectively known as net devices. In Unix and Linux you refer to these net devices by names such as eth0. 🐧
In ns-3 the net device abstraction covers both the software driver and the simulated hardware. A net device is “installed” in a Node in order to enable the Node to communicate with other Nodes in the simulation via Channels. Just as in a real computer, a Node may be connected to more than one Channel via multiple NetDevices. 🔗
The net device abstraction is represented in C++ by the class NetDevice. The NetDevice class provides methods for managing connections to Node and Channel objects; and may be specialized by developers in the object-oriented programming sense. We will use the several specialized versions of the NetDevice called CsmaNetDevice, PointToPointNetDevice, and WifiNetDevice in this tutorial. Just as an Ethernet NIC is designed to work with an Ethernet network, the CsmaNetDevice is designed to work with a CsmaChannel; the PointToPointNetDevice is designed to work with a PointToPointChannel and a WifiNetNevice is designed to work with a WifiChannel. 🛠️
Topology Helpers 🗺️
In a real network, you will find host computers with added (or built-in) NICs. In ns-3 we would say that you will find Nodes with attached NetDevices. In a large simulated network you will need to arrange many connections between Nodes, NetDevices and Channels. 🕸️✨
Since connecting NetDevices to Nodes, NetDevices to Channels, assigning IP addresses, etc., are all such common tasks in ns-3, we provide what we call topology helpers to make this as easy as possible. 💅 Like, think about it bestie — it can take a whole pile of distinct ns-3 core operations to create a NetDevice, add a MAC address, install that net device on a Node, configure the node's protocol stack, and then connect the NetDevice to a Channel. And the ops required to connect multiple devices onto multipoint channels and then wire individual networks together into internetworks? even more. 😭 So we hand you topology helper objects that bundle all those many distinct operations into one easy-to-use model, for your convenience. 🫶
✨ A First ns-3 Script ✨
If you downloaded the system as was suggested above, you'll have a release of ns-3 sitting in a directory called repos under your home directory. cd into that release directory and you should find a directory structure something like this:
AUTHORS examples scratch utils waf.bat*
bindings LICENSE src utils.py waf-tools
build ns3 test.py* utils.pyc wscript
CHANGES.html README testpy-output VERSION wutils.py
doc RELEASE_NOTES testpy.supp waf* wutils.pycChange into the examples/tutorial directory. You should see a file named first.cc chillin' there. 👑 This is a script that will create a simple point-to-point link between two nodes and echo a single packet between the nodes. Let's take a look at that script line by line, so go ahead and open first.cc in your favorite editor. 🦋
🎀 Boilerplate 🎀
The first line in the file is an emacs mode line. This tells emacs about the formatting conventions (coding style) we use in our source code.
/* -*- Mode:C++; c-file-style:"gnu"; indent-tabs-mode:nil; -*- */This is always a somewhat controversial subject, so we might as well get it out of the way immediately. 😅 The ns-3 project, like most large projects, has adopted a coding style to which all contributed code must adhere. If you want to contribute your code to the project, you will eventually have to conform to the ns-3 coding standard as described in the file doc/codingstd.txt or shown on the project web page.
We recommend that you, well, just get used to the look and feel of ns-3 code and adopt this standard whenever you are working with our code. All of the development team and contributors have done so with various amounts of grumbling. 😤 The emacs mode line above makes it easier to get the formatting correct if you use the emacs editor.
The ns-3 simulator is licensed using the GNU General Public License. You will see the appropriate GNU legalese at the head of every file in the ns-3 distribution. Often you will see a copyright notice for one of the institutions involved in the ns-3 project above the GPL text and an author listed below.
/*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License version 2 as
* published by the Free Software Foundation;
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
*/🌈 Module Includes 🌈
The code proper starts with a number of include statements.
#include "ns3/core-module.h"
#include "ns3/network-module.h"
#include "ns3/internet-module.h"
#include "ns3/point-to-point-module.h"
#include "ns3/applications-module.h"To help our high-level script users deal with the large number of include files present in the system, we group includes according to relatively large modules. 🧚 We provide a single include file that will recursively load all of the include files used in each module. Rather than having to look up exactly what header you need, and possibly have to get a number of dependencies right, we give you the ability to load a group of files at a large granularity. Is it the most efficient approach? No. 💀 Does it make writing scripts so much easier? Absolutely. periodt.
Each of the ns-3 include files is placed in a directory called ns3 (under the build directory) during the build process to help avoid include file name collisions. The ns3/core-module.h file corresponds to the ns-3 module you will find in the directory src/core in your downloaded release distribution. If you list this directory you will find a large number of header files. When you do a build, Waf will place public header files in an ns3 directory under the appropriate build/debug or build/optimized directory depending on your configuration. Waf will also automatically generate a module include file to load all of the public header files. ✨
Since you are, of course, following this tutorial religiously 🙏, you will already have done a
$ ./waf -d debug --enable-examples --enable-tests configurein order to configure the project to perform debug builds that include examples and tests. You will also have done a
$ ./wafto build the project. So now if you look in the directory ../../build/debug/ns3 you will find the four module include files shown above. You can take a look at the contents of these files and find that they do include all of the public include files in their respective modules. 👑
💜 Ns3 Namespace 💜
The next line in the first.cc script is a namespace declaration.
using namespace ns3;The ns-3 project is implemented in a C++ namespace called ns3. This groups all ns-3-related declarations in a scope outside the global namespace, which we hope will help with integration with other code. The C++ using statement introduces the ns-3 namespace into the current (global) declarative region. This is a fancy way of saying that after this declaration, you will not have to type ns3:: scope resolution operator before all of the ns-3 code in order to use it. 😌 If you are unfamiliar with namespaces, please consult almost any C++ tutorial and compare the ns3 namespace and usage here with instances of the std namespace and the using namespace std; statements you will often find in discussions of cout and streams.
🔥 Logging 🔥
The next line of the script is the following,
NS_LOG_COMPONENT_DEFINE ("FirstScriptExample");We will use this statement as a convenient place to talk about our Doxygen documentation system. If you look at the project web site, ns-3 project, you will find a link to "Documentation" in the navigation bar. If you select this link, you will be taken to our documentation page. There is a link to "Latest Release" that will take you to the documentation for the latest stable release of ns-3. If you select the "API Documentation" link, you will be taken to the ns-3 API documentation page. 🌸
Along the left side, you will find a graphical representation of the structure of the documentation. A good place to start is the NS-3 Modules "book" in the ns-3 navigation tree. If you expand Modules you will see a list of ns-3 module documentation. The concept of module here ties directly into the module include files discussed above. The ns-3 logging subsystem is discussed in the UsingLogging section, so we'll get to it later in this tutorial, but you can find out about the above statement by looking at the Core module, then expanding the Debugging tools book, and then selecting the Logging page. Click on Logging.
You should now be looking at the Doxygen documentation for the Logging module. In the list of Macro's at the top of the page you will see the entry for NS_LOG_COMPONENT_DEFINE. Before jumping in, it would probably be good to look for the "Detailed Description" of the logging module to get a feel for the overall operation. You can either scroll down or select the "More…" link under the collaboration diagram to do this.
Once you have a general idea of what is going on, go ahead and take a look at the specific NS_LOG_COMPONENT_DEFINE documentation. I won't duplicate the documentation here, but to summarize: this line declares a logging component called FirstScriptExample that allows you to enable and disable console message logging by reference to the name. 💅
⭐ Main Function ⭐
The next lines of the script you will find are,
int
main (int argc, char *argv[])
{This is just the declaration of the main function of your program (script). Just as in any C++ program, you need to define a main function that will be the first function run. There is nothing at all special here. Your ns-3 script is just a C++ program. 🫶
The next line sets the time resolution to one nanosecond, which happens to be the default value:
Time::SetResolution (Time::NS);The resolution is the smallest time value that can be represented (as well as the smallest representable difference between two time values). You can change the resolution exactly once. ⚠️ The mechanism enabling this flexibility is somewhat memory hungry, so once the resolution has been set explicitly we release the memory, preventing further updates. (If you don't set the resolution explicitly, it will default to one nanosecond, and the memory will be released when the simulation starts.)
The next two lines of the script are used to enable two logging components that are built into the Echo Client and Echo Server applications:
LogComponentEnable("UdpEchoClientApplication", LOG_LEVEL_INFO);
LogComponentEnable("UdpEchoServerApplication", LOG_LEVEL_INFO);If you have read over the Logging component documentation you will have seen that there are a number of levels of logging verbosity/detail that you can enable on each component. These two lines of code enable debug logging at the INFO level for echo clients and servers. This will result in the application printing out messages as packets are sent and received during the simulation. 🦋
Now we will get directly to the business of creating a topology and running a simulation. We use the topology helper objects to make this job as easy as possible. 😌
🍓 Topology Helpers 🍓
💖 NodeContainer 💖
The next two lines of code in our script will actually create the ns-3 Node objects that will represent the computers in the simulation.
NodeContainer nodes;
nodes.Create (2);okay so before we keep going, let's go find the docs for the NodeContainer class. Another way to get into the documentation for a given class is via the Classes tab in the Doxygen pages. If you still have the Doxygen handy, just scroll up to the top of the page and select the Classes tab. You should see a new set of tabs appear, one of which is Class List. Under that tab you will see a list of all of the ns-3 classes. Scroll down, looking for ns3::NodeContainer. When you find the class, go ahead and select it to go to the documentation for the class. ✨
You may recall that one of our key abstractions is the Node. This represents a computer to which we are going to add things like protocol stacks, applications and peripheral cards. The NodeContainer topology helper provides a convenient way to create, manage and access any Node objects that we create in order to run a simulation. The first line above just declares a NodeContainer which we call nodes. The second line calls the Create method on the nodes object and asks the container to create two nodes. As described in the Doxygen, the container calls down into the ns-3 system proper to create two Node objects and stores pointers to those objects internally. 💅
The nodes as they stand in the script do nothing. The next step in constructing a topology is to connect our nodes together into a network. The simplest form of network we support is a single point-to-point link between two nodes. We’ll construct one of those links here. 🦋
✨ PointToPointHelper ✨
We are constructing a point to point link, and, in a pattern which will become quite familiar to you, we use a topology helper object to do the low-level work required to put the link together. Recall that two of our key abstractions are the NetDevice and the Channel. In the real world, these terms correspond roughly to peripheral cards and network cables. Typically these two things are intimately tied together and one cannot expect to interchange, for example, Ethernet devices and wireless channels. Our Topology Helpers follow this intimate coupling and therefore you will use a single PointToPointHelper to configure and connect ns-3 PointToPointNetDevice and PointToPointChannel objects in this script. 🔌
The next three lines in the script are,
PointToPointHelper pointToPoint;
pointToPoint.SetDeviceAttribute ("DataRate", StringValue ("5Mbps"));
pointToPoint.SetChannelAttribute ("Delay", StringValue ("2ms"));The first line,
PointToPointHelper pointToPoint;instantiates a PointToPointHelper object on the stack. From a high-level perspective the next line,
pointToPoint.SetDeviceAttribute ("DataRate", StringValue ("5Mbps"));tells the PointToPointHelper object to use the value “5Mbps” (five megabits per second) as the “DataRate” when it creates a PointToPointNetDevice object. 🚀
From a more detailed perspective, the string “DataRate” corresponds to what we call an Attribute of the PointToPointNetDevice. If you look at the Doxygen for class ns3::PointToPointNetDevice and find the documentation for the GetTypeId method, you will find a list of Attributes defined for the device. Among these is the “DataRate” Attribute. Most user-visible ns-3 objects have similar lists of Attributes. We use this mechanism to easily configure simulations without recompiling as you will see in a following section. ✨
Similar to the “DataRate” on the PointToPointNetDevice you will find a “Delay” Attribute associated with the PointToPointChannel. The final line,
pointToPoint.SetChannelAttribute ("Delay", StringValue ("2ms"));tells the PointToPointHelper to use the value “2ms” (two milliseconds) as the value of the propagation delay of every point to point channel it subsequently creates. ⏱️
✨ NetDeviceContainer ✨
At this point in the script, we have a NodeContainer that contains two nodes. We have a PointToPointHelper that is primed and ready to make PointToPointNetDevices and wire PointToPointChannel objects between them. Just as we used the NodeContainer topology helper object to create the Nodes for our simulation, we will ask the PointToPointHelper to do the work involved in creating, configuring and installing our devices for us. We will need to have a list of all of the NetDevice objects that are created, so we use a NetDeviceContainer to hold them just as we used a NodeContainer to hold the nodes we created. The following two lines of code,
NetDeviceContainer devices;
devices = pointToPoint.Install (nodes);will finish configuring the devices and channel. The first line declares the device container mentioned above and the second does the heavy lifting. The Install method of the PointToPointHelper takes a NodeContainer as a parameter. Internally, a NetDeviceContainer is created. For each node in the NodeContainer (there must be exactly two for a point-to-point link) a PointToPointNetDevice is created and saved in the device container. A PointToPointChannel is created and the two PointToPointNetDevices are attached. When objects are created by the PointToPointHelper, the Attributes previously set in the helper are used to initialize the corresponding Attributes in the created objects. 🔧
After executing the pointToPoint.Install (nodes) call we will have two nodes, each with an installed point-to-point net device and a single point-to-point channel between them. Both devices will be configured to transmit data at five megabits per second over the channel which has a two millisecond transmission delay. 💫
✨ InternetStackHelper ✨
We now have nodes and devices configured, but we don’t have any protocol stacks installed on our nodes. The next two lines of code will take care of that.
InternetStackHelper stack;
stack.Install (nodes);The InternetStackHelper is a topology helper that is to internet stacks what the PointToPointHelper is to point-to-point net devices. The Install method takes a NodeContainer as a parameter. When it is executed, it will install an Internet Stack (TCP, UDP, IP, etc.) on each of the nodes in the node container. 🌐
✨ Ipv4AddressHelper ✨
Next we need to associate the devices on our nodes with IP addresses. We provide a topology helper to manage the allocation of IP addresses. The only user-visible API is to set the base IP address and network mask to use when performing the actual address allocation (which is done at a lower level inside the helper).
The next two lines of code in our example script, first.cc,
Ipv4AddressHelper address;
address.SetBase ("10.1.1.0", "255.255.255.0");declare an address helper object and tell it that it should begin allocating IP addresses from the network 10.1.1.0 using the mask 255.255.255.0 to define the allocatable bits. By default the addresses allocated will start at one and increase monotonically, so the first address allocated from this base will be 10.1.1.1, followed by 10.1.1.2, etc.
The low level ns-3 system actually remembers all of the IP addresses allocated and will generate a fatal error if you accidentally cause the same address to be generated twice (which is a very hard to debug error, by the way). 😱
The next line of code,
Ipv4InterfaceContainer interfaces = address.Assign (devices);performs the actual address assignment. In ns-3 we make the association between an IP address and a device using an Ipv4Interface object. Just as we sometimes need a list of net devices created by a helper for future reference we sometimes need a list of Ipv4Interface objects. The Ipv4InterfaceContainer provides this functionality. 📦
Now we have a point-to-point network built, with stacks installed and IP addresses assigned. What we need at this point are applications to generate traffic. 🎉
✨ Applications ✨
Another one of the core abstractions of the ns-3 system is the Application. In this script we use two specializations of the core ns-3 class Application called UdpEchoServerApplication and UdpEchoClientApplication. Just as we have in our previous explanations, we use helper objects to help configure and manage the underlying objects. Here, we use UdpEchoServerHelper and UdpEchoClientHelper objects to make our lives easier. 💖
✨ UdpEchoServerHelper ✨
The following lines of code in our example script, first.cc, are used to set up a UDP echo server application on one of the nodes we have previously created.
UdpEchoServerHelper echoServer (9);
ApplicationContainer serverApps = echoServer.Install (nodes.Get (1));
serverApps.Start (Seconds (1.0));
serverApps.Stop (Seconds (10.0));The first line of code in the above snippet declares the UdpEchoServerHelper. As usual, this isn’t the application itself, it is an object used to help us create the actual applications. One of our conventions is to place required Attributes in the helper constructor. In this case, the helper can’t do anything useful unless it is provided with a port number that the client also knows about. Rather than just picking one and hoping it all works out, we require the port number as a parameter to the constructor. The constructor, in turn, simply does a SetAttribute with the passed value. If you want, you can set the “Port” Attribute to another value later using SetAttribute. 🔑
Similar to many other helper objects, the UdpEchoServerHelper object has an Install method. It is the execution of this method that actually causes the underlying echo server application to be instantiated and attached to a node. Interestingly, the Install method takes a NodeContainter as a parameter just as the other Install methods we have seen. This is actually what is passed to the method even though it doesn’t look so in this case. There is a C++ implicit conversion at work here that takes the result of nodes.Get (1) (which returns a smart pointer to a node object — Ptr<Node>) and uses that in a constructor for an unnamed NodeContainer that is then passed to Install. If you are ever at a loss to find a particular method signature in C++ code that compiles and runs just fine, look for these kinds of implicit conversions. 🤯
We now see that echoServer.Install is going to install a UdpEchoServerApplication on the node found at index number one of the NodeContainer we used to manage our nodes. Install will return a container that holds pointers to all of the applications (one in this case since we passed a NodeContainer containing one node) created by the helper. 🫶
okay so Applications need a time to "start" making traffic, and they might take an optional time to "stop". We provide both, bestie. 💖
These times get set with the ApplicationContainer methods Start and Stop. They take Time parameters. In this case we use an explicit C++ conversion sequence to take the C++ double 1.0 and convert it to an ns-3 Time object using a Seconds cast. Lowkey be aware though — the conversion rules may be controlled by the model author, and C++ has its own rules, so you can't always just assume that parameters will get happily converted for you. 🙈
The two lines,
serverApps.Start (Seconds (1.0));
serverApps.Stop (Seconds (10.0));will cause the echo server application to Start (enable itself) at one second into the simulation and to Stop (disable itself) at ten seconds into the simulation. And by virtue of the fact that we declared a simulation event (the application stop event) to be executed at ten seconds, the simulation will last at least ten seconds. periodt. ⭐
UdpEchoClientHelper 🦋
The echo client application is set up in a method that's substantially similar to the one for the server. There's an underlying UdpEchoClientApplication that gets managed by an UdpEchoClientHelper.
UdpEchoClientHelper echoClient (interfaces.GetAddress (1), 9);
echoClient.SetAttribute ("MaxPackets", UintegerValue (1));
echoClient.SetAttribute ("Interval", TimeValue (Seconds (1.0)));
echoClient.SetAttribute ("PacketSize", UintegerValue (1024));
ApplicationContainer clientApps = echoClient.Install (nodes.Get (0));
clientApps.Start (Seconds (2.0));
clientApps.Stop (Seconds (10.0));For the echo client, though, we need to set five different Attributes. The first two Attributes get set during construction of the UdpEchoClientHelper. We pass parameters that are used (internally, inside the helper) to set the "RemoteAddress" and "RemotePort" Attributes, in accordance with our convention of making required Attributes into parameters in the helper constructors.
Recall that we used an Ipv4InterfaceContainer to keep track of the IP addresses we assigned to our devices. The zeroth interface in the interfaces container is gonna correspond to the IP address of the zeroth node in the nodes container. The first interface in the interfaces container corresponds to the IP address of the first node in the nodes container. So, in the first line of code (from above), we're creating the helper and telling it to set the remote address of the client to be the IP address assigned to the node the server resides on. We also tell it to arrange to send packets to port nine. 💅
The "MaxPackets" Attribute tells the client the maximum number of packets we allow it to send during the simulation. The "Interval" Attribute tells the client how long to wait between packets, and the "PacketSize" Attribute tells the client how large its packet payloads should be. With this particular combo of Attributes, we're telling the client to send one 1024-byte packet.
Just like in the case of the echo server, we tell the echo client to Start and Stop — but here we start the client one second after the server is enabled (so at two seconds into the simulation).
Simulator 👑
What we need to do at this point is to actually run the simulation. This is done using the global function Simulator::Run.
Simulator::Run ();When we previously called the methods,
serverApps.Start (Seconds (1.0));
serverApps.Stop (Seconds (10.0));
...
clientApps.Start (Seconds (2.0));
clientApps.Stop (Seconds (10.0));we actually scheduled events in the simulator at 1.0 seconds, 2.0 seconds, and two events at 10.0 seconds. When Simulator::Run is called, the system will begin looking through the list of scheduled events and executing them. First it runs the event at 1.0 seconds, which will enable the echo server application (this event may, in turn, schedule many other events). Then it runs the event scheduled for t=2.0 seconds, which will start the echo client application. Again, this event may schedule many more events. The start event implementation in the echo client application will begin the data transfer phase of the simulation by sending a packet to the server.
The act of sending the packet to the server will trigger a chain of events that get automatically scheduled behind the scenes, and those will perform the mechanics of the packet echo according to the various timing parameters we set in the script. ✨
Eventually — since we only send one packet (recall the MaxPackets Attribute was set to one) — the chain of events triggered by that single client echo request will taper off and the simulation will go idle. Once that happens, the remaining events are the Stop events for the server and the client. When these events get executed, there are no further events to process and Simulator::Run returns. The simulation is then complete. 🎀
All that remains is to clean up. This is done by calling the global function Simulator::Destroy. As the helper functions (or low level ns-3 code) executed, they arranged it so that hooks were inserted in the simulator to destroy all of the objects that were created. You did not have to keep track of any of these objects yourself — all you had to do was call Simulator::Destroy and exit. The ns-3 system took care of the hard part for you. 🫶 The remaining lines of our first ns-3 script, first.cc, do just that:
Simulator::Destroy ();
return 0;
}When the simulator will stop? 💀
ns-3 is a Discrete Event (DE) simulator. In such a simulator, each event is associated with its execution time, and the simulation proceeds by executing events in the temporal order of simulation time. Events may cause future events to be scheduled (for example, a timer may reschedule itself to expire at the next interval).
The initial events are usually triggered by each object, e.g., IPv6 will schedule Router Advertisements, Neighbor Solicitations, etc., an Application schedules the first packet sending event, etc.
When an event is processed, it may generate zero, one or more events. As a simulation executes, events are consumed, but more events may (or may not) be generated. The simulation will stop automatically when no further events are in the event queue, or when a special Stop event is found. The Stop event is created through the Simulator::Stop (stopTime); function.
There's a typical case where Simulator::Stop is absolutely necessary to stop the simulation: when there's a self-sustaining event. Self-sustaining (or recurring) events are events that always reschedule themselves. As a consequence, they always keep the event queue non-empty.
There are many protocols and modules containing recurring events, e.g.:
- FlowMonitor - periodic check for lost packets
- RIPng - periodic broadcast of routing tables update
- etc.
In these cases, Simulator::Stop is necessary to gracefully stop the simulation. In addition, when ns-3 is in emulation mode, the RealtimeSimulator is used to keep the simulation clock aligned with the machine clock, and Simulator::Stop is necessary to stop the process.
Many of the simulation programs in the tutorial do not explicitly call Simulator::Stop, since the event queue will automatically run out of events. However, these programs will also accept a call to Simulator::Stop. For example, the following additional statement in the first example program will schedule an explicit stop at 11 seconds:
+ Simulator::Stop (Seconds (11.0));
Simulator::Run ();
Simulator::Destroy ();
return 0;
}The above will not actually change the behavior of this program, since this particular simulation naturally ends after 10 seconds. But if you were to change the stop time in the above statement from 11 seconds to 1 second, you would notice that the simulation stops before any output is printed to the screen (since the output occurs around time 2 seconds of simulation time). 😭
It is important to callSimulator::Stopbefore callingSimulator::Run; otherwise,Simulator::Runmay never return control to the main program to execute the stop!
Building Your Script 🔥
We've made it trivial to build your simple scripts. All you have to do is drop your script into the scratch directory and it will automatically be built if you run Waf. Let's try it. Copy examples/tutorial/first.cc into the scratch directory after changing back into the top level directory.
$ cd ../..
$ cp examples/tutorial/first.cc scratch/myfirst.ccNow build your first example script using waf:
$ ./wafYou should see messages reporting that your myfirst example was built successfully.
Waf: Entering directory `/home/craigdo/repos/ns-3-allinone/ns-3-dev/build'
[614/708] cxx: scratch/myfirst.cc -> build/debug/scratch/myfirst_3.o
[706/708] cxx_link: build/debug/scratch/myfirst_3.o -> build/debug/scratch/myfirst
Waf: Leaving directory `/home/craigdo/repos/ns-3-allinone/ns-3-dev/build'
'build' finished successfully (2.357s)You can now run the example (note that if you build your program in the scratch directory you must run it out of the scratch directory):
$ ./waf --run scratch/myfirstYou should see some output:
Waf: Entering directory `/home/craigdo/repos/ns-3-allinone/ns-3-dev/build'
Waf: Leaving directory `/home/craigdo/repos/ns-3-allinone/ns-3-dev/build'
'build' finished successfully (0.418s)
Sent 1024 bytes to 10.1.1.2
Received 1024 bytes from 10.1.1.1
Received 1024 bytes from 10.1.1.2Here you see that the build system checks to make sure the file has been built and then runs it. You see the logging component on the echo client indicate that it has sent one 1024 byte packet to the Echo Server on 10.1.1.2. You also see the logging component on the echo server say that it has received the 1024 bytes from 10.1.1.1. The echo server silently echoes the packet and you see the echo client log that it has received its packet back from the server. 🌸
Ns-3 Source Code 📚
Now that you've used some of the ns-3 helpers, you may want to have a look at some of the source code that implements that functionality. The most recent code can be browsed on our web server at the following link: https://gitlab.com/nsnam/ns-3-dev.git. There, you'll see the Git/GitLab summary page for our ns-3 development tree.
At the top of the page, you'll see a number of links,
summary | shortlog | changelog | graph | tags | filesGo ahead and select the files link. This is what the top-level of most of our repositories will look:
drwxr-xr-x [up]
drwxr-xr-x bindings python files
drwxr-xr-x doc files
drwxr-xr-x examples files
drwxr-xr-x ns3 files
drwxr-xr-x scratch files
drwxr-xr-x src files
drwxr-xr-x utils files
-rw-r--r-- 2009-07-01 12:47 +0200 560 .hgignore file | revisions | annotate
-rw-r--r-- 2009-07-01 12:47 +0200 1886 .hgtags file | revisions | annotate
-rw-r--r-- 2009-07-01 12:47 +0200 1276 AUTHORS file | revisions | annotate
-rw-r--r-- 2009-07-01 12:47 +0200 30961 CHANGES.html file | revisions | annotate
-rw-r--r-- 2009-07-01 12:47 +0200 17987 LICENSE file | revisions | annotate
-rw-r--r-- 2009-07-01 12:47 +0200 3742 README file | revisions | annotate
-rw-r--r-- 2009-07-01 12:47 +0200 16171 RELEASE_NOTES file | revisions | annotate
-rw-r--r-- 2009-07-01 12:47 +0200 6 VERSION file | revisions | annotate
-rwxr-xr-x 2009-07-01 12:47 +0200 88110 waf file | revisions | annotate
-rwxr-xr-x 2009-07-01 12:47 +0200 28 waf.bat file | revisions | annotate
-rw-r--r-- 2009-07-01 12:47 +0200 35395 wscript file | revisions | annotate
-rw-r--r-- 2009-07-01 12:47 +0200 7673 wutils.py file | revisions | annotateOkay so our example scripts live in the examples directory, bestie. ✨ Click on examples and boom — a whole list of subdirectories. One of the files chillin' in the tutorial subdirectory is first.cc. Click on first.cc and you'll find the exact code you just walked through. Iconic. 🎀
The source code is mostly in the src directory. You can peek at source code either by clicking the directory name or by clicking the files link sitting to the right of the directory name. Click src → you get the listing of the src subdirectories. Then click core → you get a list of files. The first file you'll find (as of this writing) is abort.h. Click that abort.h link → you land on the source file for abort.h, which has useful macros for exiting scripts if abnormal conditions are detected. 💅
The source code for the helpers we've been using in this chapter? It's in the src/applications/helper directory. So go poke around the directory tree fr fr, get a feel for what's there and for the whole style of ns-3 programs. 🦋
✨ TWEAKING ✨
🫶 Using the Logging Module 🫶
We already took a lil peek at the ns-3 logging module while going over the first.cc script. Now we're gonna zoom in and see what kind of use-cases the logging subsystem was actually designed to cover. 👑
🌸 Logging Overview 🌸
So many large systems support some kind of message logging facility, and ns-3 is not an exception. Sometimes only error messages get logged to the "operator console" (which is typically stderr on Unix-based systems). In other systems, warning messages may get output too, plus more detailed informational messages. And sometimes logging facilities are used to output debug messages, which can quickly turn the output into a blur. 😭
ns-3 takes the view that all of these verbosity levels are useful, and we provide a selectable, multi-level approach to message logging. Logging can be:
- disabled completely,
- enabled on a component-by-component basis,
- or enabled globally,
and it provides selectable verbosity levels. The ns-3 log module gives you a straightforward, relatively easy-to-use way to get useful information out of your simulation. 🍓
Heads up bestie: we do provide a general purpose mechanism — tracing — to get data out of your models, and that one should be preferred for simulation output (see the tutorial section Using the Tracing System for more details on our tracing system). Logging should be preferred for debugging information, warnings, error messages, or any time you want to easily get a quick message out of your scripts or models. 💖
There are currently seven levels of log messages of increasing verbosity defined in the system.
LOG_ERROR— Log error messages (associated macro:NS_LOG_ERROR);LOG_WARN— Log warning messages (associated macro:NS_LOG_WARN);LOG_DEBUG— Log relatively rare, ad-hoc debugging messages (associated macro:NS_LOG_DEBUG);LOG_INFO— Log informational messages about program progress (associated macro:NS_LOG_INFO);LOG_FUNCTION— Log a message describing each function called (two associated macros:NS_LOG_FUNCTION, used for member functions, andNS_LOG_FUNCTION_NOARGS, used for static functions);LOG_LOGIC– Log messages describing logical flow within a function (associated macro:NS_LOG_LOGIC);LOG_ALL— Log everything mentioned above (no associated macro).
For each LOG_TYPE there is also LOG_LEVEL_TYPE that, if used, enables logging of all the levels above it in addition to it's level. (As a consequence of this, LOG_ERROR and LOG_LEVEL_ERROR and also LOG_ALL and LOG_LEVEL_ALL are functionally equivalent.) So like, enabling LOG_INFO will only enable messages provided by the NS_LOG_INFO macro, while enabling LOG_LEVEL_INFO will also enable messages provided by the NS_LOG_DEBUG, NS_LOG_WARN and NS_LOG_ERROR macros. 🧠✨
We also provide an unconditional logging macro that is always displayed, irrespective of logging levels or component selection. 💀
NS_LOG_UNCOND– Log the associated message unconditionally (no associated log level).
Each level can be requested singly or cumulatively; and logging can be set up using a shell environment variable (NS_LOG) or by logging system function call. As was seen earlier in the tutorial, the logging system has Doxygen documentation — and ngl now would be a good time to peruse the Logging Module documentation if you haven't done so. 📚
So now that you've read the documentation in great detail (slay), let's use some of that knowledge to get some interesting information out of the scratch/myfirst.cc example script you already built. 👑
💅 Enabling Logging 💅
Let's use the NS_LOG environment variable to turn on some more logging. But first, just to get our bearings, go ahead and run the last script just as you did previously:
$ ./waf --run scratch/myfirstYou should see the now-familiar output of the first ns-3 example program:
$ Waf: Entering directory `/home/craigdo/repos/ns-3-allinone/ns-3-dev/build'
Waf: Leaving directory `/home/craigdo/repos/ns-3-allinone/ns-3-dev/build'
'build' finished successfully (0.413s)
Sent 1024 bytes to 10.1.1.2
Received 1024 bytes from 10.1.1.1
Received 1024 bytes from 10.1.1.2Plot twist bestie: those "Sent" and "Received" messages you see above are actually logging messages from the UdpEchoClientApplication and UdpEchoServerApplication. 🤯 We can ask the client application, for example, to print more information by setting its logging level via the NS_LOG environment variable.
I'm gonna assume from here on that you're using an sh-like shell that uses the VARIABLE=value syntax. If you're on a csh-like shell, then you'll have to convert my examples to the setenv VARIABLE value syntax required by those shells. 🙈
Right now, the UDP echo client application is responding to the following line of code in scratch/myfirst.cc:
LogComponentEnable("UdpEchoClientApplication", LOG_LEVEL_INFO);This line of code enables the LOG_LEVEL_INFO level of logging. When we pass a logging level flag, we're actually enabling the given level and all lower levels. In this case, we've enabled NS_LOG_INFO, NS_LOG_DEBUG, NS_LOG_WARN and NS_LOG_ERROR. And here's the cute part — we can increase the logging level and get more information without changing the script and recompiling, just by setting the NS_LOG environment variable like this:
$ export NS_LOG=UdpEchoClientApplication=level_allThis sets the shell environment variable NS_LOG to the string:
UdpEchoClientApplication=level_allThe left hand side of the assignment is the name of the logging component we want to set, and the right hand side is the flag we want to use. In this case, we're turning on all of the debugging levels for the application. If you run the script with NS_LOG set this way, the ns-3 logging system will pick up the change and you should see the following output:
Waf: Entering directory `/home/craigdo/repos/ns-3-allinone/ns-3-dev/build'
Waf: Leaving directory `/home/craigdo/repos/ns-3-allinone/ns-3-dev/build'
'build' finished successfully (0.404s)
UdpEchoClientApplication:UdpEchoClient()
UdpEchoClientApplication:SetDataSize(1024)
UdpEchoClientApplication:StartApplication()
UdpEchoClientApplication:ScheduleTransmit()
UdpEchoClientApplication:Send()
Sent 1024 bytes to 10.1.1.2
Received 1024 bytes from 10.1.1.1
UdpEchoClientApplication:HandleRead(0x6241e0, 0x624a20)
Received 1024 bytes from 10.1.1.2
UdpEchoClientApplication:StopApplication()
UdpEchoClientApplication:DoDispose()
UdpEchoClientApplication:~UdpEchoClient()The extra debug info the application is giving us comes from the NS_LOG_FUNCTION level. It shows every single time a function in the application is called during script execution. Generally, using (at least) NS_LOG_FUNCTION(this) in member functions is preferred. Use NS_LOG_FUNCTION_NOARGS() only in static functions. But note, however, that there are no requirements in the ns-3 system that models must support any particular logging functionality. The decision about how much information gets logged is left to the individual model developer. In the case of the echo applications, a good deal of log output is available. 🎀
You can now see a log of the function calls that were made to the application. And if you look closely, you'll notice a single colon</> between the string UdpEchoClientApplication and the method name, where you might have expected a C++ scope operator (::). That is intentional. 👀
The name is not actually a class name — it's a logging component name. When there's a one-to-one correspondence between a source file and a class, this will generally be the class name, but you should understand that it is not actually a class name, and there's a single colon there instead of a double colon to remind you in a relatively subtle way to conceptually separate the logging component name from the class name. 🦋💖
Okay so sometimes it's genuinely hard to figure out which method is actually spitting out a log message. Like, you're staring at the text above and you're lowkey wondering where the string "Received 1024 bytes from 10.1.1.2" even comes from?? Bestie, you can resolve this by OR-ing the prefix_func level into the NS_LOG environment variable. Try doing the following:
$ export 'NS_LOG=UdpEchoClientApplication=level_all|prefix_func'⚠️ Note that the quotes are required, since the vertical bar we use to indicate an OR operation is also a Unix pipe connector. The more you know ✨
Now if you run the script, you'll see the logging system makes sure that every message from the given log component is prefixed with the component name. 🌸
Waf: Entering directory `/home/craigdo/repos/ns-3-allinone/ns-3-dev/build'
Waf: Leaving directory `/home/craigdo/repos/ns-3-allinone/ns-3-dev/build'
'build' finished successfully (0.417s)
UdpEchoClientApplication:UdpEchoClient()
UdpEchoClientApplication:SetDataSize(1024)
UdpEchoClientApplication:StartApplication()
UdpEchoClientApplication:ScheduleTransmit()
UdpEchoClientApplication:Send()
UdpEchoClientApplication:Send(): Sent 1024 bytes to 10.1.1.2
Received 1024 bytes from 10.1.1.1
UdpEchoClientApplication:HandleRead(0x6241e0, 0x624a20)
UdpEchoClientApplication:HandleRead(): Received 1024 bytes from 10.1.1.2
UdpEchoClientApplication:StopApplication()
UdpEchoClientApplication:DoDispose()
UdpEchoClientApplication:~UdpEchoClient()You can now see all of the messages coming from the UDP echo client application are identified as such. The message "Received 1024 bytes from 10.1.1.2" is now clearly identified as coming from the echo client application. The remaining message must be coming from the UDP echo server application. We can enable that component by entering a colon-separated list of components in the NS_LOG environment variable. 🫶
$ export 'NS_LOG=UdpEchoClientApplication=level_all|prefix_func:
UdpEchoServerApplication=level_all|prefix_func'Warning: You will need to remove the newline after the : in the example text above, which is only there for document formatting purposes. Don't copy it, bestie, I'm begging 😭
Now if you run the script, you'll see all of the log messages from both the echo client and server applications. You may see that this can be very useful in debugging problems. 🔥
Waf: Entering directory `/home/craigdo/repos/ns-3-allinone/ns-3-dev/build'
Waf: Leaving directory `/home/craigdo/repos/ns-3-allinone/ns-3-dev/build'
'build' finished successfully (0.406s)
UdpEchoServerApplication:UdpEchoServer()
UdpEchoClientApplication:UdpEchoClient()
UdpEchoClientApplication:SetDataSize(1024)
UdpEchoServerApplication:StartApplication()
UdpEchoClientApplication:StartApplication()
UdpEchoClientApplication:ScheduleTransmit()
UdpEchoClientApplication:Send()
UdpEchoClientApplication:Send(): Sent 1024 bytes to 10.1.1.2
UdpEchoServerApplication:HandleRead(): Received 1024 bytes from 10.1.1.1
UdpEchoServerApplication:HandleRead(): Echoing packet
UdpEchoClientApplication:HandleRead(0x624920, 0x625160)
UdpEchoClientApplication:HandleRead(): Received 1024 bytes from 10.1.1.2
UdpEchoServerApplication:StopApplication()
UdpEchoClientApplication:StopApplication()
UdpEchoClientApplication:DoDispose()
UdpEchoServerApplication:DoDispose()
UdpEchoClientApplication:~UdpEchoClient()
UdpEchoServerApplication:~UdpEchoServer()It's also sometimes useful to be able to see the simulation time at which a log message is generated. You can do this by ORing in the prefix_time bit. ⭐
$ export 'NS_LOG=UdpEchoClientApplication=level_all|prefix_func|prefix_time:
UdpEchoServerApplication=level_all|prefix_func|prefix_time'Again, you'll have to remove the newline above. If you run the script now, you should see the following output:
Waf: Entering directory `/home/craigdo/repos/ns-3-allinone/ns-3-dev/build'
Waf: Leaving directory `/home/craigdo/repos/ns-3-allinone/ns-3-dev/build'
'build' finished successfully (0.418s)
0s UdpEchoServerApplication:UdpEchoServer()
0s UdpEchoClientApplication:UdpEchoClient()
0s UdpEchoClientApplication:SetDataSize(1024)
1s UdpEchoServerApplication:StartApplication()
2s UdpEchoClientApplication:StartApplication()
2s UdpEchoClientApplication:ScheduleTransmit()
2s UdpEchoClientApplication:Send()
2s UdpEchoClientApplication:Send(): Sent 1024 bytes to 10.1.1.2
2.00369s UdpEchoServerApplication:HandleRead(): Received 1024 bytes from 10.1.1.1
2.00369s UdpEchoServerApplication:HandleRead(): Echoing packet
2.00737s UdpEchoClientApplication:HandleRead(0x624290, 0x624ad0)
2.00737s UdpEchoClientApplication:HandleRead(): Received 1024 bytes from 10.1.1.2
10s UdpEchoServerApplication:StopApplication()
10s UdpEchoClientApplication:StopApplication()
UdpEchoClientApplication:DoDispose()
UdpEchoServerApplication:DoDispose()
UdpEchoClientApplication:~UdpEchoClient()
UdpEchoServerApplication:~UdpEchoServer()You can see that the constructor for the UdpEchoServer was called at a simulation time of 0 seconds. This is actually happening before the simulation starts, but the time is displayed as zero seconds. The same is true for the UdpEchoClient constructor message. 🙈
Recall that the scratch/first.cc script started the echo server application at one second into the simulation. You can now see that the StartApplication method of the server is, in fact, called at one second. You can also see that the echo client application is started at a simulation time of two seconds as we requested in the script. 👑
You can now follow the progress of the simulation from the ScheduleTransmit call in the client that calls Send to the HandleRead callback in the echo server application. Note that the elapsed time for the packet to be sent across the point-to-point link is 3.69 milliseconds. You see the echo server logging a message telling you that it has echoed the packet, and then, after another channel delay, you see the echo client receive the echoed packet in its HandleRead method. ✨
There's a lot that's happening under the covers in this simulation that you're not seeing as well. You can very easily follow the entire process by turning on all of the logging components in the system. Try setting the NS_LOG variable to the following:
$ export 'NS_LOG=*=level_all|prefix_func|prefix_time'The asterisk above is the logging component wildcard. This will turn on all of the logging in all of the components used in the simulation. I won't reproduce the output here (as of this writing it produces 1265 lines of output for the single packet echo) but you can redirect this information into a file and look through it with your favorite editor if you like:
$ ./waf --run scratch/myfirst > log.out 2>&1I personally use this extremely verbose version of logging when I'm presented with a problem and I have no idea where things are going wrong. I can follow the progress of the code quite easily without having to set breakpoints and step through code in a debugger. I can just edit up the output in my favorite editor and search around for things I expect, and see things happening that I don't expect. When I have a general idea about what's going wrong, I transition into a debugger for a fine-grained examination of the problem. This kind of output can be especially useful when your script does something completely unexpected. If you're stepping using a debugger you may miss an unexpected excursion completely. Logging the excursion makes it quickly visible. 💜
🎀 Adding Logging to your Code 🎀
You can add new logging to your simulations by making calls to the log component via several macros. Let's do so in the myfirst.cc script we have in the scratch directory. 🌸
Recall that we have defined a logging component in that script:
NS_LOG_COMPONENT_DEFINE ("FirstScriptExample");You now know that you can enable all of the logging for this component by setting the NS_LOG environment variable to the various levels. Let's go ahead and add some logging to the script. The macro used to add an informational level log message is NS_LOG_INFO. Go ahead and add one (just before we start creating the nodes) that tells you that the script is "Creating Topology." This is done as in this code snippet:
Open scratch/myfirst.cc in your favorite editor and add the line:
NS_LOG_INFO ("Creating Topology");right before the lines:
NodeContainer nodes;
nodes.Create (2);Now build the script using waf and clear the NS_LOG variable to turn off the torrent of logging we previously enabled:
$ ./waf
$ export NS_LOG=Now, if you run the script:
$ ./waf --run scratch/myfirstyou will not see your new message, since its associated logging component (FirstScriptExample) has not been enabled. In order to see your message you'll have to enable the FirstScriptExample logging component with a level greater than or equal to NS_LOG_INFO. If you just want to see this particular level of logging, you can enable it by:
$ export NS_LOG=FirstScriptExample=infoIf you now run the script you'll see your new "Creating Topology" log message: 🦋
Waf: Entering directory `/home/craigdo/repos/ns-3-allinone/ns-3-dev/build'
Waf: Leaving directory `/home/craigdo/repos/ns-3-allinone/ns-3-dev/build'
'build' finished successfully (0.404s)
Creating Topology
Sent 1024 bytes to 10.1.1.2
Received 1024 bytes from 10.1.1.1
Received 1024 bytes from 10.1.1.2🌈 Using Command Line Arguments 🌈
💅 Overriding Default Attributes 💅
Another way you can change how ns-3 scripts behave without editing and building is via command line arguments. We provide a mechanism to parse command line arguments and automatically set local and global variables based on those arguments. 🔥
The first step in using the command line argument system is to declare the command line parser. This is done quite simply (in your main program) as in the following code:
int
main (int argc, char *argv[])
{
...
CommandLine cmd;
cmd.Parse (argc, argv);
...
}This simple two-line snippet is actually very useful by itself. It opens the door to the ns-3 global variable and Attribute systems. Go ahead and add those two lines of code to the scratch/myfirst.cc script at the start of main. Go ahead and build the script and run it, but ask the script for help in the following way:
$ ./waf --run "scratch/myfirst --PrintHelp"This is gonna ask Waf to run the scratch/myfirst script and yeet the command line argument --PrintHelp at it 💌 The quotes are required bestie, they're literally the only way to sort out which program gets which argument — no cap. The command line parser will now see the --PrintHelp argument and respond with:
Waf: Entering directory `/home/craigdo/repos/ns-3-allinone/ns-3-dev/build'
Waf: Leaving directory `/home/craigdo/repos/ns-3-allinone/ns-3-dev/build'
'build' finished successfully (0.413s)
TcpL4Protocol:TcpStateMachine()
CommandLine:HandleArgument(): Handle arg name=PrintHelp value=
--PrintHelp: Print this help message.
--PrintGroups: Print the list of groups.
--PrintTypeIds: Print all TypeIds.
--PrintGroup=[group]: Print all TypeIds of group.
--PrintAttributes=[typeid]: Print all attributes of typeid.
--PrintGlobals: Print the list of globals.✨ Let's get into the Attribute lore ✨
Okay so let's hyperfixate on the --PrintAttributes option for a sec 🫶 We already hinted at the ns-3 Attribute system when we were walking through the first.cc script. Remember these lines of code?
PointToPointHelper pointToPoint;
pointToPoint.SetDeviceAttribute ("DataRate", StringValue ("5Mbps"));
pointToPoint.SetChannelAttribute ("Delay", StringValue ("2ms"));And we mentioned that DataRate was actually an Attribute of the PointToPointNetDevice. So let's use the command line argument parser to peek at the Attributes of the PointToPointNetDevice 🔍 The help listing says we should provide a TypeId. That corresponds to the class name of the class the Attributes belong to. In this case it's gonna be ns3::PointToPointNetDevice. So let's just type:
$ ./waf --run "scratch/myfirst --PrintAttributes=ns3::PointToPointNetDevice"The system will print out all of the Attributes of this kind of net device. Among the Attributes you'll see listed is:
--ns3::PointToPointNetDevice::DataRate=[32768bps]:
The default data rate for point to point linksThis is the default value that gets used when a PointToPointNetDevice is created in the system. We overrode this default with the Attribute setting in the PointToPointHelper above. So let's use the default values for the point-to-point devices and channels by deleting the SetDeviceAttribute call and the SetChannelAttribute call from the myfirst.cc we have in the scratch directory.
Your script should now just declare the PointToPointHelper and not do any set operations, like in this example:
...
NodeContainer nodes;
nodes.Create (2);
PointToPointHelper pointToPoint;
NetDeviceContainer devices;
devices = pointToPoint.Install (nodes);
...Go ahead and build the new script with Waf (./waf) and let's go back and enable some logging from the UDP echo server application and turn on the time prefix 🌸
$ export 'NS_LOG=UdpEchoServerApplication=level_all|prefix_time'If you run the script, you should now see the following output:
Waf: Entering directory `/home/craigdo/repos/ns-3-allinone/ns-3-dev/build'
Waf: Leaving directory `/home/craigdo/repos/ns-3-allinone/ns-3-dev/build'
'build' finished successfully (0.405s)
0s UdpEchoServerApplication:UdpEchoServer()
1s UdpEchoServerApplication:StartApplication()
Sent 1024 bytes to 10.1.1.2
2.25732s Received 1024 bytes from 10.1.1.1
2.25732s Echoing packet
Received 1024 bytes from 10.1.1.2
10s UdpEchoServerApplication:StopApplication()
UdpEchoServerApplication:DoDispose()
UdpEchoServerApplication:~UdpEchoServer()Okay so remember the last time we looked at the simulation time at which the packet was received by the echo server, it was at 2.00369 seconds.
2.00369s UdpEchoServerApplication:HandleRead(): Received 1024 bytes from 10.1.1.1Now it's receiving the packet at 2.25732 seconds. Why tho? Because we literally just dropped the data rate of the PointToPointNetDevice down to its default of 32768 bits per second from five megabits per second 💀
If we were to provide a new DataRate using the command line, we could speed our simulation up again. We do this in the following way, according to the formula implied by the help item:
$ ./waf --run "scratch/myfirst --ns3::PointToPointNetDevice::DataRate=5Mbps"This will set the default value of the DataRate Attribute back to five megabits per second. Are you surprised by the result? 👀 It turns out that in order to get the original behavior of the script back, we will have to set the speed-of-light delay of the channel as well. We can ask the command line system to print out the Attributes of the channel just like we did for the net device:
$ ./waf --run "scratch/myfirst --PrintAttributes=ns3::PointToPointChannel"We discover the Delay Attribute of the channel is set in the following way:
--ns3::PointToPointChannel::Delay=[0ns]:
Transmission delay through the channelWe can then set both of these default values through the command line system:
$ ./waf --run "scratch/myfirst
--ns3::PointToPointNetDevice::DataRate=5Mbps
--ns3::PointToPointChannel::Delay=2ms"in which case we recover the timing we had when we explicitly set the DataRate and Delay in the script:
Waf: Entering directory `/home/craigdo/repos/ns-3-allinone/ns-3-dev/build'
Waf: Leaving directory `/home/craigdo/repos/ns-3-allinone/ns-3-dev/build'
'build' finished successfully (0.417s)
0s UdpEchoServerApplication:UdpEchoServer()
1s UdpEchoServerApplication:StartApplication()
Sent 1024 bytes to 10.1.1.2
2.00369s Received 1024 bytes from 10.1.1.1
2.00369s Echoing packet
Received 1024 bytes from 10.1.1.2
10s UdpEchoServerApplication:StopApplication()
UdpEchoServerApplication:DoDispose()
UdpEchoServerApplication:~UdpEchoServer()Note that the packet is again received by the server at 2.00369 seconds ✨ We could actually set any of the Attributes used in the script this way. In particular we could set the UdpEchoClient Attribute MaxPackets to some other value than one.
How would you go about that? 🤔 Give it a try! Remember you have to comment out the place we override the default Attribute and explicitly set MaxPackets in the script. Then you have to rebuild the script. You'll also have to find the syntax for actually setting the new default attribute value using the command line help facility. Once you have this figured out you should be able to control the number of packets echoed from the command line 🎀 Since we're nice folks, we'll tell you that your command line should end up looking something like:
$ ./waf --run "scratch/myfirst
--ns3::PointToPointNetDevice::DataRate=5Mbps
--ns3::PointToPointChannel::Delay=2ms
--ns3::UdpEchoClient::MaxPackets=2"A natural question to arise at this point is how do I even learn about the existence of all of these attributes? Again, the command line help facility has a feature for this. If we ask for command line help we should see:
$ ./waf --run "scratch/myfirst --PrintHelp"
myfirst [Program Arguments] [General Arguments]
General Arguments:
--PrintGlobals: Print the list of globals.
--PrintGroups: Print the list of groups.
--PrintGroup=[group]: Print all TypeIds of group.
--PrintTypeIds: Print all TypeIds.
--PrintAttributes=[typeid]: Print all attributes of typeid.
--PrintHelp: Print this help message.If you select the "PrintGroups" argument, you should see a list of all registered TypeId groups 🌈 The group names are aligned with the module names in the source directory (although with a leading capital letter). Printing out all of the information at once would be way too much, so a further filter is available to print information on a per-group basis. So, focusing again on the point-to-point module:
./waf --run "scratch/myfirst --PrintGroup=PointToPoint"
TypeIds in group PointToPoint:
ns3::PointToPointChannel
ns3::PointToPointNetDevice
ns3::PointToPointRemoteChannel
ns3::PppHeaderand from here, one can find the possible TypeId names to search for attributes, such as in the --PrintAttributes=ns3::PointToPointChannel example shown above 👑
Another way to find out about attributes is through the ns-3 Doxygen; there's a page that lists out all of the registered attributes in the simulator 🦋
🎀 Hooking Your Own Values 🎀
You can also add your own hooks to the command line system 💅 This is done quite simply by using the AddValue method to the command line parser.
Let's use this facility to specify the number of packets to echo in a completely different way. Let's add a local variable called nPackets to the main function. We'll initialize it to one to match our previous default behavior. To allow the command line parser to change this value, we need to hook the value into the parser. We do this by adding a call to AddValue. Go ahead and change the scratch/myfirst.cc script to start with the following code:
int
main (int argc, char *argv[])
{
uint32_t nPackets = 1;
CommandLine cmd;
cmd.AddValue("nPackets", "Number of packets to echo", nPackets);
cmd.Parse (argc, argv);
...Scroll down to the point in the script where we set the MaxPackets Attribute and change it so that it's set to the variable nPackets instead of the constant 1, as shown below.
echoClient.SetAttribute ("MaxPackets", UintegerValue (nPackets));Now if you run the script and provide the --PrintHelp argument, you should see your new User Argument listed in the help display 👀
Try,
$ ./waf --run "scratch/myfirst --PrintHelp"
Waf: Entering directory `/home/craigdo/repos/ns-3-allinone/ns-3-dev/build'
Waf: Leaving directory `/home/craigdo/repos/ns-3-allinone/ns-3-dev/build'
'build' finished successfully (0.403s)
--PrintHelp: Print this help message.
--PrintGroups: Print the list of groups.
--PrintTypeIds: Print all TypeIds.
--PrintGroup=[group]: Print all TypeIds of group.
--PrintAttributes=[typeid]: Print all attributes of typeid.
--PrintGlobals: Print the list of globals.
User Arguments:
--nPackets: Number of packets to echoIf you want to specify the number of packets to echo, you can now do so by setting the --nPackets argument in the command line,
$ ./waf --run "scratch/myfirst --nPackets=2"You should now see
Waf: Entering directory `/home/craigdo/repos/ns-3-allinone/ns-3-dev/build'
Waf: Leaving directory `/home/craigdo/repos/ns-3-allinone/ns-3-dev/build'
'build' finished successfully (0.404s)
0s UdpEchoServerApplication:UdpEchoServer()
1s UdpEchoServerApplication:StartApplication()
Sent 1024 bytes to 10.1.1.2
2.25732s Received 1024 bytes from 10.1.1.1
2.25732s Echoing packet
Received 1024 bytes from 10.1.1.2
Sent 1024 bytes to 10.1.1.2
3.25732s Received 1024 bytes from 10.1.1.1
3.25732s Echoing packet
Received 1024 bytes from 10.1.1.2
10s UdpEchoServerApplication:StopApplication()
UdpEchoServerApplication:DoDispose()
UdpEchoServerApplication:~UdpEchoServer()BESTIE. YOU ECHOED TWO PACKETS. 💖 Two whole packets. Look at you go. That's not a drill, that's a real live simulation doing a real live thing, and it's literally not even hard, right? 🎀
So the vibe check here is this: if you're just an ns-3 user, you can use the command line argument system to boss around global values and Attributes. ✨ If you're a model author, you can add brand new Attributes to your Objects and they'll just… automatically be settable by your users through the command line system. No extra wiring. Slay. 👑 And if you're a script author, you can add new variables to your scripts and hook them into the command line system pretty much painlessly. 💅
🌈 Using the Tracing System 🌈
Okay so like, the whole point of simulation is to generate output so you can go study it later. That's the entire assignment. 📝 And the ns-3 tracing system is the main character of that whole operation. Since ns-3 is a C++ program, technically you could just use the standard C++ facilities for spitting out output:
#include <iostream>
...
int main ()
{
...
std::cout << "The value of x is " << x << std::endl;
...
}You could even pull in the logging module to add a lil structure to your solution. 🙈 But there are, like, so many well-known problems that come with that approach, so ns-3 went ahead and gave us a generic event tracing subsystem to handle the issues they thought were important. 💀
The basic goals of the ns-3 tracing system are:
- For basic tasks, the tracing system should allow the user to generate standard tracing for popular tracing sources, and to customize which objects generate the tracing;
- Intermediate users must be able to extend the tracing system to modify the output format generated, or to insert new tracing sources, without modifying the core of the simulator;
- Advanced users can modify the simulator core to add new tracing sources and sinks.
The ns-3 tracing system is built on the concepts of independent tracing sources and tracing sinks, plus a uniform mechanism for connecting sources to sinks. 🦋 Trace sources are entities that can signal events that happen in a simulation and give you access to interesting underlying data. For example, a trace source could tell you when a packet is received by a net device and hand over the packet contents for any interested trace sinks.
But here's the tea ☕ — trace sources are not useful by themselves. They have to be "connected" to other pieces of code that actually do something useful with the info the sink provides. Trace sinks are the consumers of the events and data that the trace sources hand out. For example, you could make a trace sink that would (when connected to the trace source from the previous example) print out the interesting parts of the received packet. 🫶
The rationale for this explicit division is to let users attach new types of sinks to existing tracing sources without requiring edits and recompilation of the core of the simulator. So in the example above, a user could define a new tracing sink in her script and attach it to an existing tracing source defined in the simulation core by editing only the user script. Iconic behavior, honestly. ✨
In this tutorial, we're gonna walk through some pre-defined sources and sinks and show how they can be customized with very little user effort. 💅 See the ns-3 manual or how-to sections for info on advanced tracing configuration, including extending the tracing namespace and creating new tracing sources.
🌸 ASCII Tracing 🌸
ns-3 provides helper functionality that wraps the low-level tracing system so you don't have to deal with all the fiddly details involved in configuring some easily understood packet traces. If you enable this functionality, you'll see output in ASCII files — hence the name. 🙈 For those familiar with ns-2 output, this type of trace is analogous to the out.tr generated by many scripts.
Let's just jump right in and add some ASCII tracing output to our scratch/myfirst.cc script. 🎀 Right before the call to Simulator::Run (), add the following lines of code:
AsciiTraceHelper ascii;
pointToPoint.EnableAsciiAll (ascii.CreateFileStream ("myfirst.tr"));Like in many other ns-3 idioms, this code uses a helper object to help create ASCII traces. The second line has two nested method calls in it. The "inside" method, CreateFileStream(), uses an unnamed object idiom to create a file stream object on the stack (with no object name) and passes it down to the method being called. We'll go into this more later, but all you need to know at this point is that you're creating an object representing a file named myfirst.tr and passing it into ns-3. 🌈 You're telling ns-3 to deal with the lifetime issues of the created object, and also to deal with problems caused by a little-known (intentional) limitation of C++ ofstream objects relating to copy constructors. 💀
The outside call, to EnableAsciiAll(), tells the helper that you want ASCII tracing enabled on all point-to-point devices in your simulation, and that you want the (provided) trace sinks to write out info about packet movement in ASCII format. ✨
For those familiar with ns-2, the traced events are equivalent to the popular trace points that log +, -, d, and r events.
You can now build the script and run it from the command line:
$ ./waf --run scratch/myfirstJust like you've seen a million times before, you'll get some messages from Waf and then 'build' finished successfully with some number of messages from the running program. 🫶
When it ran, the program will have created a file named myfirst.tr. But because of the way Waf works, the file is not created in the local directory — it gets created at the top-level directory of the repository by default. 😭 If you want to control where the traces get saved, you can use the --cwd option of Waf to specify it. We didn't do that, so we need to change into the top level directory of our repo and take a look at the ASCII trace file myfirst.tr in your favorite editor. 💅
🦋 Parsing Ascii Traces 🦋
There's a lot of info in there in a pretty dense form, ngl, but the first thing to notice is that there are a number of distinct lines in this file. It might be hard to see that clearly unless you widen your window considerably. 👀
Each line in the file corresponds to a trace event. In this case, we're tracing events on the transmit queue present in every point-to-point net device in the simulation. The transmit queue is a queue that every packet destined for a point-to-point channel must pass through. Note that each line in the trace file begins with a lone character (there's a space after it). That character means the following:
+: An enqueue operation occurred on the device queue;-: A dequeue operation occurred on the device queue;d: A packet was dropped, typically because the queue was full;r: A packet was received by the net device.
Let's take a more detailed look at the first line in the trace file. I'll break it down into sections (indented for clarity) with a reference number on the left side:
+
2
/NodeList/0/DeviceList/0/$ns3::PointToPointNetDevice/TxQueue/Enqueue
ns3::PppHeader (
Point-to-Point Protocol: IP (0x0021))
ns3::Ipv4Header (
tos 0x0 ttl 64 id 0 protocol 17 offset 0 flags [none]
length: 1052 10.1.1.1 > 10.1.1.2)
ns3::UdpHeader (
length: 1032 49153 > 9)
Payload (size=1024)The first section of this expanded trace event (reference number 0) is the operation. We've got a + character, so this corresponds to an enqueue operation on the transmit queue. The second section (reference 1) is the simulation time expressed in seconds. You may recall that we asked the UdpEchoClientApplication to start sending packets at two seconds. Here we see confirmation that this is, indeed, happening. ✨
The next section of the example trace (reference 2) tells us which trace source originated this event (expressed in the tracing namespace). You can think of the tracing namespace kinda like a filesystem namespace. 📁 The root of the namespace is the NodeList. This corresponds to a container managed in the ns-3 core code that holds all of the nodes that get created in a script. Just like a filesystem can have directories under the root, we can have node numbers in the NodeList. So the string /NodeList/0 refers to the zeroth node in the NodeList, which we typically think of as "node 0". In each node there's a list of devices that have been installed. That list appears next in the namespace. You can see that this trace event comes from DeviceList/0, which is the zeroth device installed in the node. 💅
The next string, $ns3::PointToPointNetDevice, tells you what kind of device is in the zeroth position of the device list for node zero. Recall that the operation + found at reference 00 meant that an enqueue operation happened on the transmit queue of the device. That's reflected in the final segments of the "trace path", which are TxQueue/Enqueue. 🔥
The remaining sections in the trace should be pretty intuitive. References 3-4 indicate that the packet is encapsulated in the point-to-point protocol. References 5-7 show that the packet has an IP version four header and has originated from IP address 10.1.1.1 and is destined for 10.1.1.2. References 8-9 show that this packet has a UDP header and, finally, reference 10 shows that the payload is the expected 1024 bytes. 🌸
The next line in the trace file shows the same packet being dequeued from the transmit queue on the same node.
The third line in the trace file shows the packet being received by the net device on the node with the echo server. I've reproduced that event below. 👇
r
2.25732
/NodeList/1/DeviceList/0/$ns3::PointToPointNetDevice/MacRx
ns3::Ipv4Header (
tos 0x0 ttl 64 id 0 protocol 17 offset 0 flags [none]
length: 1052 10.1.1.1 > 10.1.1.2)
ns3::UdpHeader (
length: 1032 49153 > 9)
Payload (size=1024)Okay so bestie, notice that the trace operation is now r and the simulation time has jumped to 2.25732 seconds? 💅 If you've been following along like a good little coder, that means you left the DataRate of the net devices and the channel Delay at their default values. No cap. That time should look familiar because you've literally seen it before in a previous section. 👑
The trace source namespace entry (reference 02) has changed to show that this event is coming from node 1 (/NodeList/1) and the packet reception trace source (/MacRx). ✨ It should be super easy for you to follow the packet's whole journey through the topology just by looking at the rest of the traces in the file. Periodt.
🎀 PCAP Tracing 🎀
The ns-3 device helpers can also create trace files in the .pcap format. 🎀 The acronym pcap (usually lowercase) stands for packet capture, and it's actually an API that includes the definition of a .pcap file format. The most popular program that can read and display this format is Wireshark (formerly called Ethereal). But lowkey, there are tons of traffic trace analyzers that use this packet format. We encourage you to exploit all the many tools available for analyzing pcap traces. In this tutorial, we're focusing on viewing pcap traces with tcpdump. 💖
The code to enable pcap tracing is literally a one-liner:
pointToPoint.EnablePcapAll ("myfirst");Go ahead and insert this line right after the ASCII tracing code we just added to scratch/myfirst.cc. 🫶 Notice that we only passed the string “myfirst,” and not “myfirst.pcap” or something similar. That's because the parameter is a prefix, not a complete file name. The helper will actually create a trace file for every point-to-point device in the simulation. The file names get built using the prefix, the node number, the device number, and a “.pcap” suffix. No cap.
In our example script, we'll eventually see files named “myfirst-0-0.pcap” and “myfirst-1-0.pcap” — those are the pcap traces for node 0-device 0 and node 1-device 0, respectively. 💅
Once you've added that line to enable pcap tracing, you can run the script in the usual way:
$ ./waf --run scratch/myfirstIf you look at the top level directory of your distribution, you should now see three log files: myfirst.tr is the ASCII trace file we already examined. myfirst-0-0.pcap and myfirst-1-0.pcap are the new pcap files we just generated. ✨
📖 Reading output with tcpdump 📖
The easiest thing to do at this point is to use tcpdump to look at the pcap files. 👑
$ tcpdump -nn -tt -r myfirst-0-0.pcap
reading from file myfirst-0-0.pcap, link-type PPP (PPP)
2.000000 IP 10.1.1.1.49153 > 10.1.1.2.9: UDP, length 1024
2.514648 IP 10.1.1.2.9 > 10.1.1.1.49153: UDP, length 1024
tcpdump -nn -tt -r myfirst-1-0.pcap
reading from file myfirst-1-0.pcap, link-type PPP (PPP)
2.257324 IP 10.1.1.1.49153 > 10.1.1.2.9: UDP, length 1024
2.257324 IP 10.1.1.2.9 > 10.1.1.1.49153: UDP, length 1024You can see in the dump of myfirst-0-0.pcap (the client device) that the echo packet is sent at 2 seconds into the simulation. If you look at the second dump (myfirst-1-0.pcap), you can see that packet being received at 2.257324 seconds. You see the packet being echoed back at 2.257324 seconds in the second dump, and finally, you see the packet being received back at the client in the first dump at 2.514648 seconds. 🦋
🦋 Reading output with Wireshark 🦋
If you're unfamiliar with Wireshark, there's a web site where you can download programs and documentation: http://www.wireshark.org/. 🌐
Wireshark is a graphical user interface that can display these trace files. If you have Wireshark available, you can open each of the trace files and display the contents as if you had captured the packets using a packet sniffer. 💖
🏗️ BUILDING TOPOLOGIES 🏗️
🚌 Building a Bus Network Topology 🚌
In this section, we're gonna expand our mastery of ns-3 network devices and channels to cover an example of a bus network. 🚌 ns-3 provides a net device and channel we call CSMA (Carrier Sense Multiple Access).
The ns-3 CSMA device models a simple network in the spirit of Ethernet. A real Ethernet uses CSMA/CD (Carrier Sense Multiple Access with Collision Detection) with exponentially increasing backoff to contend for the shared transmission medium. The ns-3 CSMA device and channel models only a subset of this. No cap.
Just like we saw point-to-point topology helper objects when constructing point-to-point topologies, we'll see equivalent CSMA topology helpers in this section. The appearance and operation of these helpers should look super familiar to you. 💅
We provide an example script in our examples/tutorial directory. This script builds on the first.cc script and adds a CSMA network to the point-to-point simulation we've already considered. Go ahead and open examples/tutorial/second.cc in your favorite editor. 👩💻 You've already seen enough ns-3 code to understand most of what's going on in this example, but we'll go over the entire script and examine some of the output.
Just like in the first.cc example (and in all ns-3 examples), the file begins with an emacs mode line and some GPL boilerplate. 📜
The actual code begins by loading module include files just like in the first.cc example:
#include "ns3/core-module.h"
#include "ns3/network-module.h"
#include "ns3/csma-module.h"
#include "ns3/internet-module.h"
#include "ns3/point-to-point-module.h"
#include "ns3/applications-module.h"
#include "ns3/ipv4-global-routing-helper.h"One thing that can be surprisingly useful is a small bit of ASCII art that shows a cartoon of the network topology constructed in the example. You'll find a similar “drawing” in most of our examples. 🎨
In this case, you can see that we're going to extend our point-to-point example (the link between nodes n0 and n1 below) by hanging a bus network off the right side. Notice that this is the default network topology since you can actually vary the number of nodes created on the LAN. If you set nCsma to one, there will be a total of two nodes on the LAN (CSMA channel) — one required node and one “extra” node. By default there are three “extra” nodes as seen below:
// Default Network Topology
//
// 10.1.1.0
// n0 -------------- n1 n2 n3 n4
// point-to-point | | | |
// ================
// LAN 10.1.2.0Then the ns-3 namespace is used and a logging component is defined. This is all just like it was in first.cc, so nothing new yet. ✨
using namespace ns3;
NS_LOG_COMPONENT_DEFINE ("SecondScriptExample");The main program begins with a slightly different twist. We use a verbose flag to determine whether or not the UdpEchoClientApplication and UdpEchoServerApplication logging components are enabled. This flag defaults to true (the logging components are enabled) but lets us turn off logging during regression testing of this example. 🧪
You'll see some familiar code that lets you change the number of devices on the CSMA network via command line argument. We did something similar when we allowed the number of packets sent to be changed in the section on command line arguments. The last line makes sure you have at least one “extra” node. 💅
The code consists of variations of previously covered API so you should be entirely comfortable with the following code at this point in the tutorial. 😌
bool verbose = true;
uint32_t nCsma = 3;
CommandLine cmd;
cmd.AddValue ("nCsma", "Number of \"extra\" CSMA nodes/devices", nCsma);
cmd.AddValue ("verbose", "Tell echo applications to log if true", verbose);
cmd.Parse (argc, argv);
if (verbose)
{
LogComponentEnable("UdpEchoClientApplication", LOG_LEVEL_INFO);
LogComponentEnable("UdpEchoServerApplication", LOG_LEVEL_INFO);
}
nCsma = nCsma == 0 ? 1 : nCsma;The next step is to create two nodes that we'll connect via the point-to-point link. The NodeContainer is used to do this just like in first.cc. 🫶
NodeContainer p2pNodes;
p2pNodes.Create (2);Next, we declare another NodeContainer to hold the nodes that will be part of the bus (CSMA) network. First, we just instantiate the container object itself. ✨
NodeContainer csmaNodes;
csmaNodes.Add (p2pNodes.Get (1));
csmaNodes.Create (nCsma);The next line of code Gets the first node (as in having an index of one) from the point-to-point node container and adds it to the container of nodes that will get CSMA devices. The node in question is going to end up with a point-to-point device and a CSMA device. We then create a number of “extra” nodes that compose the remainder of the CSMA network. Since we already have one node in the CSMA network – the one that will have both a point-to-point and CSMA net device, the number of “extra” nodes means the number nodes you desire in the CSMA section minus one. 🧮
The next bit of code should be quite familiar by now. We instantiate a PointToPointHelper and set the associated default Attributes so that we create a five megabit per second transmitter on devices created using the helper and a two millisecond delay on channels created by the helper. 💅
PointToPointHelper pointToPoint;
pointToPoint.SetDeviceAttribute ("DataRate", StringValue ("5Mbps"));
pointToPoint.SetChannelAttribute ("Delay", StringValue ("2ms"));
NetDeviceContainer p2pDevices;
p2pDevices = pointToPoint.Install (p2pNodes);We then instantiate a NetDeviceContainer to keep track of the point-to-point net devices and we Install devices on the point-to-point nodes. ✨
We mentioned above that you were going to see a helper for CSMA devices and channels, and the next lines introduce them. The CsmaHelper works just like a PointToPointHelper, but it creates and connects CSMA devices and channels. In the case of a CSMA device and channel pair, notice that the data rate is specified by a channel Attribute instead of a device Attribute. This is because a real CSMA network does not allow one to mix, for example, 10Base-T and 100Base-T devices on a given channel. We first set the data rate to 100 megabits per second, and then set the speed-of-light delay of the channel to 6560 nano-seconds (arbitrarily chosen as 1 nanosecond per foot over a 2000 meter segment). Notice that you can set an Attribute using its native data type. 🌈
CsmaHelper csma;
csma.SetChannelAttribute ("DataRate", StringValue ("100Mbps"));
csma.SetChannelAttribute ("Delay", TimeValue (NanoSeconds (6560))); NetDeviceContainer csmaDevices;
csmaDevices = csma.Install (csmaNodes);So same energy as before, different helper 💅 — we make a NetDeviceContainer to hold the devices that our CsmaHelper is about to spawn, exactly like we did for the PointToPointHelper. Then we call Install on the CsmaHelper and hand it the csmaNodes NodeContainer, and it installs those devices into every node in there. Slay. ✨
Now — vibe check 🫶 — we have nodes, we have devices, we have channels, and we have absolutely zero protocol stacks. Tragic. So just like in first.cc, we roll out the InternetStackHelper to fix that:
InternetStackHelper stack;
stack.Install (p2pNodes.Get (0));
stack.Install (csmaNodes);Real talk, remember we yoinked one node out of the p2pNodes container and dropped it into csmaNodes? That's why we only have to install stacks on the remaining p2pNodes node, plus every node in csmaNodes — and boom, that covers literally every node in the simulation. No double-stacking, no drama. 😌
Next up, IP addresses, same as the first.cc script — Ipv4AddressHelper is our girl. First we use network 10.1.1.0 to make the two addresses our two point-to-point devices need:
Ipv4AddressHelper address;
address.SetBase ("10.1.1.0", "255.255.255.0");
Ipv4InterfaceContainer p2pInterfaces;
p2pInterfaces = address.Assign (p2pDevices);And yes bestie, we're saving those interfaces in a container again, because later we're gonna want to yank addressing info back out when we set up the applications. 💅
Now the CSMA devices need IPs too. Same operation as the point-to-point case, except this time we're doing it on a container with a variable number of CSMA devices — because we made that number changeable by command line argument, remember? Iconic. In this case the CSMA devices get addresses from network number 10.1.2.0, like so:
address.SetBase ("10.1.2.0", "255.255.255.0");
Ipv4InterfaceContainer csmaInterfaces;
csmaInterfaces = address.Assign (csmaDevices);🦋 Applications, but make it internetwork 🦋
Topology: built. Applications: not built. So let's fix that. This bit is fundamentally similar to the applications section of first.cc, except we're putting the server on a node that has a CSMA device and the client on the node that only has a point-to-point device. Main character placement, honestly. 👑
First, the echo server. We create a UdpEchoServerHelper and hand the constructor a required Attribute value — the server port number. (Recall: you can change this later with SetAttribute if you want, but the constructor requires it. No skipping. 🚫)
UdpEchoServerHelper echoServer (9);
ApplicationContainer serverApps = echoServer.Install (csmaNodes.Get (nCsma));
serverApps.Start (Seconds (1.0));
serverApps.Stop (Seconds (10.0));Okay so let's think about this csmaNodes.Get (nCsma) thing because it's giving ✨index math✨. The csmaNodes NodeContainer holds one of the nodes created for the point-to-point network plus nCsma "extra" nodes. What we actually want is the last of the "extra" nodes. The zeroth entry of csmaNodes is the point-to-point node. So the easy way to think about it: if you make one "extra" CSMA node, it lands at index one of the container. By induction, if you make nCsma "extra" nodes, the last one is at index nCsma. And that's exactly what you see in the Get on the first line of code. 🧠💖
The client is set up exactly like in the first.cc script. Again we pass required Attributes to the UdpEchoClientHelper in the constructor — in this case the remote address and the port. We tell the client to send packets to the server we just installed on the last of the "extra" CSMA nodes, and we install the client on the leftmost point-to-point node from the topology illustration.
UdpEchoClientHelper echoClient (csmaInterfaces.GetAddress (nCsma), 9);
echoClient.SetAttribute ("MaxPackets", UintegerValue (1));
echoClient.SetAttribute ("Interval", TimeValue (Seconds (1.0)));
echoClient.SetAttribute ("PacketSize", UintegerValue (1024));
ApplicationContainer clientApps = echoClient.Install (p2pNodes.Get (0));
clientApps.Start (Seconds (2.0));
clientApps.Stop (Seconds (10.0));🌈 Global routing, aka the magic ✨ 🌈
Since we've genuinely built an internetwork here, we need some form of internetwork routing. ns-3 has your back with what it calls global routing. Global routing takes advantage of the fact that the whole internetwork is accessible inside the simulation and runs through all of the nodes created for the simulation — so it does the hard work of setting up routing for you, no router configuring required. Obsessed. 😭💜
Basically what happens: each node acts like it were an OSPF router that communicates instantly and magically with all the other routers behind the scenes. Each node generates link advertisements and sends them straight to a global route manager, which uses that global info to build the routing tables for every node. Setting this up is a genuine one-liner:
Ipv4GlobalRoutingHelper::PopulateRoutingTables ();🎀 pcap tracing, but with an extra parameter 🎀
Next we enable pcap tracing. The first line — pcap tracing in the point-to-point helper — should feel like an old friend by now. The second line enables pcap tracing in the CSMA helper and there's an extra parameter you haven't met yet. 👀
pointToPoint.EnablePcapAll ("second");
csma.EnablePcap ("second", csmaDevices.Get (1), true);So. The CSMA network is a multi-point-to-point network, which means there can be (and in this case, are) multiple endpoints on a shared medium. Every one of those endpoints has a net device attached. And there are two basic ways to gather trace info from a network like that:
- Create a trace file for each net device and store only the packets that device emits or consumes.
- Pick one of the devices and put it in promiscuous mode. That single device then "sniffs" the network for all packets and stores them all in a single pcap file. That's literally how
tcpdumpworks. 🕵️♀️
That final parameter? It tells the CSMA helper whether or not to arrange for packets to be captured in promiscuous mode.
In this example we're picking one device on the CSMA network and asking it to do a promiscuous sniff of the whole network — emulating what tcpdump would do. If you were on a Linux machine you might do something like tcpdump -i eth0 to get the trace. Here, we specify the device with csmaDevices.Get(1), which selects the first device in the container, and setting that final parameter to true enables promiscuous captures.
Last section of code just runs and cleans up the simulation, exactly like the first.cc example:
Simulator::Run ();
Simulator::Destroy ();
return 0;
}🌸 Running it, bestie 🌸
To run this example, copy the second.cc example script into the scratch directory and use waf to build, just like you did with first.cc. If you're sitting in the top-level directory of the repository, you just type:
$ cp examples/tutorial/second.cc scratch/mysecond.cc
$ ./wafWarning: we use the filesecond.ccas one of our regression tests to verify that it works exactly as we think it should, so your tutorial experience stays a positive one. That means an executable namedsecondalready exists in the project. To avoid any confusion about what you're actually executing, please do the renaming tomysecond.ccsuggested above. 🙈
If you're following the tutorial religiously (you are, aren't you 😌) you'll still have the NS_LOG variable set, so go ahead and clear that variable and run the program:
$ export NS_LOG=
$ ./waf --run scratch/mysecondSince we set up the UDP echo applications to log just like we did in first.cc, you'll see similar output when you run the script:
Waf: Entering directory `/home/craigdo/repos/ns-3-allinone/ns-3-dev/build'
Waf: Leaving directory `/home/craigdo/repos/ns-3-allinone/ns-3-dev/build'
'build' finished successfully (0.415s)
Sent 1024 bytes to 10.1.2.4
Received 1024 bytes from 10.1.1.1
Received 1024 bytes from 10.1.2.4Quick decode, because these three lines are doing a whole lot 💅:
- The first message,
Sent 1024 bytes to 10.1.2.4, is the UDP echo client sending a packet to the server. In this case the server is on a different network (10.1.2.0). - The second message,
Received 1024 bytes from 10.1.1.1, is from the UDP echo server, generated when it receives the echo packet. - The final message,
Received 1024 bytes from 10.1.2.4, is from the echo client, telling us it got its echo back from the server.
⭐ Trace files, aka the receipts ⭐
If you now go look in the top level directory, you'll find three trace files:
second-0-0.pcap second-1-0.pcap second-2-0.pcapLet's take a sec to appreciate the naming, because it's not random. They all have the same form: <name>-<node>-<device>.pcap. So:
second-0-0.pcapis the pcap trace from node zero, device zero. That's the point-to-point net device on node zero.second-1-0.pcapis the pcap trace for device zero on node one — also a point-to-point net device.second-2-0.pcapis the pcap trace for device zero on node two.
And if you refer back to the topology illustration at the start of the section: node zero is the leftmost node of the point-to-point link, node one is the node that has both a point-to-point device and a CSMA device, and node two is the first "extra" node on the CSMA network — and its device zero is the one we selected to capture the promiscuous-mode trace. 🎯
💀 Following the packet through the internetwork 💀
Now let's follow that echo packet through the internetwork. First, do a tcpdump of the trace file for the leftmost point-to-point node — node zero:
$ tcpdump -nn -tt -r second-0-0.pcapYou should see the contents of the pcap file displayed:
reading from file second-0-0.pcap, link-type PPP (PPP)
2.000000 IP 10.1.1.1.49153 > 10.1.2.4.9: UDP, length 1024
2.017607 IP 10.1.2.4.9 > 10.1.1.1.49153: UDP, length 1024The first line of the dump tells us the link type is PPP (point-to-point) — which is exactly what we expected. Then you see the echo packet leaving node zero via the device associated with IP address 10.1.1.1, headed for IP address 10.1.2.4 (the rightmost CSMA node). That packet will move over the point-to-point link and get received by the point-to-point net device on node one. Let's take a look 👀:
$ tcpdump -nn -tt -r second-1-0.pcapYou should now see the pcap trace output of the other side of the point-to-point link:
okay so bestie — we're looking at the trace on the other end now, and honestly? it's giving ✨storytime✨. Let's read it together.
reading from file second-1-0.pcap, link-type PPP (PPP)
2.003686 IP 10.1.1.1.49153 > 10.1.2.4.9: UDP, length 1024
2.013921 IP 10.1.2.4.9 > 10.1.1.1.49153: UDP, length 1024So here the link type is also PPP, which is exactly what we expected, no surprises 💅. You can see the packet from IP address 10.1.1.1 — the one that got sent at 2.000000 seconds — headed toward IP address 10.1.2.4 show up on this interface. And now, internally to this node, that packet is gonna get forwarded over to the CSMA interface, and we should see it pop out on that device, heading for its ultimate destination. main character packet behavior fr fr.
Remember how we picked node 2 as the promiscuous sniffer node for the CSMA network? Bestie. Let's go look at second-2-0.pcap and see if it's there 👀.
$ tcpdump -nn -tt -r second-2-0.pcapYou should now see the promiscuous dump of node two, device zero:
reading from file second-2-0.pcap, link-type EN10MB (Ethernet)
2.007698 ARP, Request who-has 10.1.2.4 (ff:ff:ff:ff:ff:ff) tell 10.1.2.1, length 50
2.007710 ARP, Reply 10.1.2.4 is-at 00:00:00:00:00:06, length 50
2.007803 IP 10.1.1.1.49153 > 10.1.2.4.9: UDP, length 1024
2.013815 ARP, Request who-has 10.1.2.1 (ff:ff:ff:ff:ff:ff) tell 10.1.2.4, length 50
2.013828 ARP, Reply 10.1.2.1 is-at 00:00:00:00:00:03, length 50
2.013921 IP 10.1.2.4.9 > 10.1.1.1.49153: UDP, length 1024See how the link type is now "Ethernet"? 🎀 But something new has appeared. The bus network needs ARP, the Address Resolution Protocol. Node one knows it needs to send the packet to IP address 10.1.2.4, but bestie — it does not know the MAC address of the corresponding node 😭. So it broadcasts on the CSMA network (ff:ff:ff:ff:ff:ff) asking for the device that has IP address 10.1.2.4. And in this case, the rightmost node replies saying it's at MAC address 00:00:00:00:00:06. Note that node two is not directly involved in this exchange, but it's sniffing the network and reporting all of the traffic it sees. nosy queen 👑 we love her.
This exchange is seen in the following lines:
2.007698 ARP, Request who-has 10.1.2.4 (ff:ff:ff:ff:ff:ff) tell 10.1.2.1, length 50
2.007710 ARP, Reply 10.1.2.4 is-at 00:00:00:00:00:06, length 50Then node one, device one goes ahead and sends the echo packet to the UDP echo server at IP address 10.1.2.4.
2.007803 IP 10.1.1.1.49153 > 10.1.2.4.9: UDP, length 1024The server receives that echo request and turns the packet around, trying to send it back to the source. The server knows this address is on another network that it reaches via IP address 10.1.2.1 — and that's because we initialized global routing and it has figured all of this out for us 💖. But the echo server node doesn't know the MAC address of the first CSMA node, so it has to ARP for it, just like the first CSMA node had to do. circular behavior, iconic honestly.
2.013815 ARP, Request who-has 10.1.2.1 (ff:ff:ff:ff:ff:ff) tell 10.1.2.4, length 50
2.013828 ARP, Reply 10.1.2.1 is-at 00:00:00:00:00:03, length 50The server then sends the echo back to the forwarding node.
2.013921 IP 10.1.2.4.9 > 10.1.1.1.49153: UDP, length 1024Now let's look back at the rightmost node of the point-to-point link:
$ tcpdump -nn -tt -r second-1-0.pcapYou can now see the echoed packet coming back onto the point-to-point link as the last line of the trace dump 🦋.
reading from file second-1-0.pcap, link-type PPP (PPP)
2.003686 IP 10.1.1.1.49153 > 10.1.2.4.9: UDP, length 1024
2.013921 IP 10.1.2.4.9 > 10.1.1.1.49153: UDP, length 1024Lastly, you can look back at the node that originated the echo:
$ tcpdump -nn -tt -r second-0-0.pcapand see that the echoed packet arrives back at the source at 2.017607 seconds ✨.
reading from file second-0-0.pcap, link-type PPP (PPP)
2.000000 IP 10.1.1.1.49153 > 10.1.2.4.9: UDP, length 1024
2.017607 IP 10.1.2.4.9 > 10.1.1.1.49153: UDP, length 1024Finally, recall that we added the ability to control the number of CSMA devices in the simulation by command line argument 🎀. You can change this argument in the same way as when we looked at changing the number of packets echoed in the first.cc example. Try running the program with the number of "extra" devices set to four:
$ ./waf --run "scratch/mysecond --nCsma=4"You should now see:
Waf: Entering directory `/home/craigdo/repos/ns-3-allinone/ns-3-dev/build'
Waf: Leaving directory `/home/craigdo/repos/ns-3-allinone/ns-3-dev/build'
'build' finished successfully (0.405s)
At time 2s client sent 1024 bytes to 10.1.2.5 port 9
At time 2.0118s server received 1024 bytes from 10.1.1.1 port 49153
At time 2.0118s server sent 1024 bytes to 10.1.1.1 port 49153
At time 2.02461s client received 1024 bytes from 10.1.2.5 port 9Notice the echo server has now been relocated to the last of the CSMA nodes, which is 10.1.2.5 instead of the default case, 10.1.2.4. she moved houses 💅.
🦋 snooping on ONE device instead of the whole party 🦋
It is possible that you may not be satisfied with a trace file generated by a bystander in the CSMA network 🙈. Like, maybe you really want a trace from a single device and you are genuinely not interested in any other traffic on the network. You can do this fairly easily, bestie.
Let's take a look at scratch/mysecond.cc and add that code enabling us to be more specific. ns-3 helpers provide methods that take a node number and device number as parameters. Go ahead and replace the EnablePcap calls with the calls below:
pointToPoint.EnablePcap ("second", p2pNodes.Get (0)->GetId (), 0);
csma.EnablePcap ("second", csmaNodes.Get (nCsma)->GetId (), 0, false);
csma.EnablePcap ("second", csmaNodes.Get (nCsma-1)->GetId (), 0, false);We know we want to create a pcap file with the base name second, and we also know the device of interest in both cases is going to be zero, so those parameters are not really interesting 🫶.
In order to get the node number, you have two choices:
- First, nodes are numbered in a monotonically increasing fashion starting from zero in the order in which you created them. One way to get a node number is to figure this number out "manually" by contemplating the order of node creation. If you take a look at the network topology illustration at the beginning of the file, we did this for you and you can see that the last CSMA node is going to be node number
nCsma + 1. This approach can become annoyingly difficult in larger simulations 😭. - An alternate way, which we use here, is to realize that the
NodeContainerscontain pointers to ns-3NodeObjects. TheNodeObject has a method calledGetIdwhich will return that node's ID, which is the node number we seek. Let's go take a look at the Doxygen for theNodeand locate that method, which is further down in the ns-3 core code than we've seen so far; but sometimes you have to search diligently for useful things 🔍.
Go to the Doxygen documentation for your release (recall that you can find it on the project web site). You can get to the Node documentation by looking through at the "Classes" tab and scrolling down the "Class List" until you find ns3::Node. Select ns3::Node and you will be taken to the documentation for the Node class. If you now scroll down to the GetId method and select it, you will be taken to the detailed documentation for the method. Using the GetId method can make determining node numbers much easier in complex topologies. slay.
Let's clear the old trace files out of the top-level directory to avoid confusion about what is going on:
$ rm *.pcap
$ rm *.trIf you build the new script and run the simulation setting nCsma to 100:
$ ./waf --run "scratch/mysecond --nCsma=100"you will see the following output:
Waf: Entering directory `/home/craigdo/repos/ns-3-allinone/ns-3-dev/build'
Waf: Leaving directory `/home/craigdo/repos/ns-3-allinone/ns-3-dev/build'
'build' finished successfully (0.407s)
At time 2s client sent 1024 bytes to 10.1.2.101 port 9
At time 2.0068s server received 1024 bytes from 10.1.1.1 port 49153
At time 2.0068s server sent 1024 bytes to 10.1.1.1 port 49153
At time 2.01761s client received 1024 bytes from 10.1.2.101 port 9Note that the echo server is now located at 10.1.2.101, which corresponds to having 100 "extra" CSMA nodes with the echo server on the last one 🎀. If you list the pcap files in the top level directory you will see:
second-0-0.pcap second-100-0.pcap second-101-0.pcapThe trace file second-0-0.pcap is the "leftmost" point-to-point device, which is the echo packet source. The file second-101-0.pcap corresponds to the rightmost CSMA device, which is where the echo server resides. You may have noticed that the final parameter on the call to enable pcap tracing on the echo server node was false. This means the trace gathered on that node was in non-promiscuous mode. not her being polite and only listening to her own business 💅.
To illustrate the difference between promiscuous and non-promiscuous traces, we also requested a non-promiscuous trace for the next-to-last node. Go ahead and take a look at the tcpdump for second-100-0.pcap.
$ tcpdump -nn -tt -r second-100-0.pcapYou can now see that node 100 is really just a bystander in the echo exchange 🙈. The only packets it receives are the ARP requests, which are broadcast to the entire CSMA network.
reading from file second-100-0.pcap, link-type EN10MB (Ethernet)
2.006698 ARP, Request who-has 10.1.2.101 (ff:ff:ff:ff:ff:ff) tell 10.1.2.1, length 50
2.013815 ARP, Request who-has 10.1.2.1 (ff:ff:ff:ff:ff:ff) tell 10.1.2.101, length 50Now take a look at the tcpdump for second-101-0.pcap.
$ tcpdump -nn -tt -r second-101-0.pcapYou can now see that node 101 is really the participant in the echo exchange. main character energy, no cap ✨👑.
reading from file second-101-0.pcap, link-type EN10MB (Ethernet)
2.006698 ARP, Request who-has 10.1.2.101 (ff:ff:ff:ff:ff:ff) tell 10.1.2.1, length 50
2.006698 ARP, Reply 10.1.2.101 is-at 00:00:00:00:00:67, length 50
2.006803 IP 10.1.1.1.49153 > 10.1.2.101.9: UDP, length 1024
2.013803 ARP, Request who-has 10.1.2.1 (ff:ff:ff:ff:ff:ff) tell 10.1.2.101, length 50
2.013828 ARP, Reply 10.1.2.1 is-at 00:00:00:00:00:03, length 50
2.013828 IP 10.1.2.101.9 > 10.1.1.1.49153: UDP, length 1024🫶 Models, Attributes and Reality 🫶
This is a convenient place to make a small excursion and make an important point. It may or may not be obvious to you, but whenever one is using a simulation, it is important to understand exactly what is being modeled and what is not. It is tempting, for example, to think of the CSMA devices and channels used in the previous section as if they were real Ethernet devices; and to expect a simulation result to directly reflect what will happen in a real Ethernet. This is not the case. 💀
bestie. a simulation is a model. it is not the thing. do not @ me.
A model is, by definition, an abstraction of reality 💭. And it's ultimately the simulation script author's job to determine the so-called “range of accuracy” and “domain of applicability” of the simulation as a whole, and therefore its constituent parts. Periodt.
In some cases, like Csma, it can be fairly easy to determine what is not modeled. By reading the model description (csma.h) you can find that there is no collision detection in the CSMA model and decide on how applicable its use will be in your simulation or what caveats you may want to include with your results. In other cases, it can be quite easy to configure behaviors that might not agree with any reality you can go out and buy. It will prove worthwhile to spend some time investigating a few such instances, and how easily you can swerve outside the bounds of reality in your simulations.
As you have seen, ns-3 provides Attributes which a user can easily set to change model behavior. Consider two of the Attributes of the CsmaNetDevice: Mtu and EncapsulationMode. The Mtu attribute indicates the Maximum Transmission Unit to the device. This is the size of the largest Protocol Data Unit (PDU) that the device can send.
The MTU defaults to 1500 bytes in the CsmaNetDevice. This default corresponds to a number found in RFC 894, “A Standard for the Transmission of IP Datagrams over Ethernet Networks.” The number is actually derived from the maximum packet size for 10Base5 (full-spec Ethernet) networks – 1518 bytes. If you subtract the DIX encapsulation overhead for Ethernet packets (18 bytes) you will end up with a maximum possible data size (MTU) of 1500 bytes. One can also find that the MTU for IEEE 802.3 networks is 1492 bytes. This is because LLC/SNAP encapsulation adds an extra eight bytes of overhead to the packet. In both cases, the underlying hardware can only send 1518 bytes, but the data size is different.
In order to set the encapsulation mode, the CsmaNetDevice provides an Attribute called EncapsulationMode which can take on the values Dix or Llc. These correspond to Ethernet and LLC/SNAP framing respectively.
If one leaves the Mtu at 1500 bytes and changes the encapsulation mode to Llc, the result will be a network that encapsulates 1500 byte PDUs with LLC/SNAP framing resulting in packets of 1526 bytes, which would be illegal in many networks, since they can transmit a maximum of 1518 bytes per packet. This would most likely result in a simulation that quite subtly does not reflect the reality you might be expecting. Yikes.
Just to complicate the picture, there exist jumbo frames (1500 < MTU <= 9000 bytes) and super-jumbo (MTU > 9000 bytes) frames that are not officially sanctioned by IEEE but are available in some high-speed (Gigabit) networks and NICs. One could leave the encapsulation mode set to Dix, and set the Mtu Attribute on a CsmaNetDevice to 64000 bytes – even though an associated CsmaChannel DataRate was set at 10 megabits per second. This would essentially model an Ethernet switch made out of vampire-tapped 1980s-style 10Base5 networks that support super-jumbo datagrams. This is certainly not something that was ever made, nor is likely to ever be made, but it is quite easy for you to configure.
In the previous example, you used the command line to create a simulation that had 100 Csma nodes. You could have just as easily created a simulation with 500 nodes. If you were actually modeling that 10Base5 vampire-tap network, the maximum length of a full-spec Ethernet cable is 500 meters, with a minimum tap spacing of 2.5 meters. That means there could only be 200 taps on a real network. You could have quite easily built an illegal network in that way as well. This may or may not result in a meaningful simulation depending on what you are trying to model.
Similar situations can occur in many places in ns-3 and in any simulator. For example, you may be able to position nodes in such a way that they occupy the same space at the same time, or you may be able to configure amplifiers or noise levels that violate the basic laws of physics.
ns-3 generally favors flexibility, and many models will allow freely setting Attributes without trying to enforce any arbitrary consistency or particular underlying spec.
The thing to take home from this is that ns-3 is going to provide a super-flexible base for you to experiment with. It is up to you to understand what you are asking the system to do and to make sure that the simulations you create have some meaning and some connection with a reality defined by you. That's the tea. ☕
✨ Building a Wireless Network Topology ✨
In this section we are going to further expand our knowledge of ns-3 network devices and channels to cover an example of a wireless network. ns-3 provides a set of 802.11 models that attempt to provide an accurate MAC-level implementation of the 802.11 specification and a “not-so-slow” PHY-level model of the 802.11a specification.
Just as we have seen both point-to-point and CSMA topology helper objects when constructing point-to-point topologies, we will see equivalent Wifi topology helpers in this section. The appearance and operation of these helpers should look quite familiar to you.
We provide an example script in our examples/tutorial directory. This script builds on the second.cc script and adds a Wi-Fi network. Go ahead and open examples/tutorial/third.cc in your favorite editor. You will have already seen enough ns-3 code to understand most of what is going on in this example, but there are a few new things, so we will go over the entire script and examine some of the output.
Just as in the second.cc example (and in all ns-3 examples) the file begins with an emacs mode line and some GPL boilerplate.
Take a look at the ASCII art (reproduced below) that shows the default network topology constructed in the example. You can see that we are going to further extend our example by hanging a wireless network off of the left side. Notice that this is a default network topology since you can actually vary the number of nodes created on the wired and wireless networks. Just as in the second.cc script case, if you change nCsma, it will give you a number of “extra” CSMA nodes. Similarly, you can set nWifi to control how many STA (station) nodes are created in the simulation. There will always be one AP (access point) node on the wireless network. By default there are three “extra” CSMA nodes and three wireless STA nodes.
The code begins by loading module include files just as was done in the second.cc example. There are a couple of new includes corresponding to the wifi module and the mobility module which we will discuss below.
#include "ns3/core-module.h"
#include "ns3/point-to-point-module.h"
#include "ns3/network-module.h"
#include "ns3/applications-module.h"
#include "ns3/wifi-module.h"
#include "ns3/mobility-module.h"
#include "ns3/csma-module.h"
#include "ns3/internet-module.h"The network topology illustration follows:
// Default Network Topology
//
// Wifi 10.1.3.0
// AP
// * * * *
// | | | | 10.1.1.0
// n5 n6 n7 n0 -------------- n1 n2 n3 n4
// point-to-point | | | |
// ================
// LAN 10.1.2.0You can see that we are adding a new network device to the node on the left side of the point-to-point link that becomes the access point for the wireless network. A number of wireless STA nodes are created to fill out the new 10.1.3.0 network as shown on the left side of the illustration.
After the illustration, the ns-3 namespace is used and a logging component is defined. This should all be quite familiar by now.
using namespace ns3;
NS_LOG_COMPONENT_DEFINE ("ThirdScriptExample");The main program begins just like second.cc by adding some command line parameters for enabling or disabling logging components and for changing the number of devices created.
bool verbose = true;
uint32_t nCsma = 3;
uint32_t nWifi = 3;
CommandLine cmd;
cmd.AddValue ("nCsma", "Number of \"extra\" CSMA nodes/devices", nCsma);
cmd.AddValue ("nWifi", "Number of wifi STA devices", nWifi);
cmd.AddValue ("verbose", "Tell echo applications to log if true", verbose);
cmd.Parse (argc,argv);
if (verbose)
{
LogComponentEnable("UdpEchoClientApplication", LOG_LEVEL_INFO);
LogComponentEnable("UdpEchoServerApplication", LOG_LEVEL_INFO);
}Just as in all of the previous examples, the next step is to create two nodes that we will connect via the point-to-point link.
NodeContainer p2pNodes;
p2pNodes.Create (2);Next, we see an old friend. We instantiate a PointToPointHelper and set the associated default Attributes so that we create a five megabit per second transmitter on devices created using the helper and a two millisecond delay on channels created by the helper. We then Install the devices on the nodes and the channel between them.
PointToPointHelper pointToPoint;
pointToPoint.SetDeviceAttribute ("DataRate", StringValue ("5Mbps"));
pointToPoint.SetChannelAttribute ("Delay", StringValue ("2ms"));
NetDeviceContainer p2pDevices;
p2pDevices = pointToPoint.Install (p2pNodes);Next, we declare another NodeContainer to hold the nodes that will be part of the bus (CSMA) network.
NodeContainer csmaNodes;
csmaNodes.Add (p2pNodes.Get (1));
csmaNodes.Create (nCsma);The next line of code Gets the first node (as in having an index of one) from the point-to-point node container and adds it to the container of nodes that will get CSMA devices. The node in question is going to end up with a point-to-point device and a CSMA device. We then create a number of “extra” nodes that compose the remainder of the CSMA network.
We then instantiate a CsmaHelper and set its Attributes as we did in the previous example. We create a NetDeviceContainer to keep track of the created CSMA net devices and then we Install CSMA devices on the selected nodes.
CsmaHelper csma;
csma.SetChannelAttribute ("DataRate", StringValue ("100Mbps"));
csma.SetChannelAttribute ("Delay", TimeValue (NanoSeconds (6560)));
NetDeviceContainer csmaDevices;
csmaDevices = csma.Install (csmaNodes);Next, we are going to create the nodes that will be part of the Wi-Fi network. We are going to create a number of “station” nodes as specified by the command line argument, and we are going to use the “leftmost” node of the point-to-point link as the node for the access point.
okay so bestie, the STA nodes are actual literal things now, we made them, and then we grab the AP node from the p2p container — it's index 0:
NodeContainer wifiStaNodes;
wifiStaNodes.Create (nWifi);
NodeContainer wifiApNode = p2pNodes.Get (0);Next up we build the wifi devices <em>and</em> the channel that links all these wifi nodes together. First, the PHY and channel helpers get configured:
YansWifiChannelHelper channel = YansWifiChannelHelper::Default ();
YansWifiPhyHelper phy = YansWifiPhyHelper::Default ();Keeping it simple, this uses the default PHY layer config and channel models — those are documented in the API doxygen for the YansWifiChannelHelper::Default and YansWifiPhyHelper::Default methods, go peek 🫶. Once those objects exist, we make a channel object and hand it to our PHY layer object manager, so that every PHY layer object the YansWifiPhyHelper makes shares the same underlying channel — same wireless medium, so they can talk to each other and interfere with each other, no cap:
phy.SetChannel (channel.Create ());Now that the PHY helper is set, we move up to the MAC layer. Here we're going with non-Qos MACs. The WifiMacHelper object is what sets MAC parameters:
WifiHelper wifi;
wifi.SetRemoteStationManager ("ns3::AarfWifiManager");
WifiMacHelper mac;The SetRemoteStationManager method tells the helper which rate control algorithm type to use. Here it's asking for the AARF algorithm — details are, of course, in Doxygen. iykyk 💅
Next: the type of MAC, the SSID of the infrastructure network we want, and making sure our stations don't do active probing:
Ssid ssid = Ssid ("ns-3-ssid");
mac.SetType ("ns3::StaWifiMac",
"Ssid", SsidValue (ssid),
"ActiveProbing", BooleanValue (false));So first this creates an 802.11 service set identifier (SSID) object — that'll be used to set the value of the "Ssid" Attribute of the MAC layer implementation. The particular kind of MAC layer the helper will create is specified by Attribute as being the "ns3::StaWifiMac" type. The "QosSupported" Attribute is set to false by default for WifiMacHelper objects. Combine those two configs and it means the MAC instance created next will be a non-QoS non-AP station (STA) in an infrastructure BSS — aka a BSS with an AP. And finally, "ActiveProbing" is set to false, which means no probe requests will be sent by MACs this helper creates. periodt.
Once all the station-specific params are fully set, both at MAC and PHY layers, we hit our now-familiar Install method to make the Wi-Fi devices for these stations:
NetDeviceContainer staDevices;
staDevices = wifi.Install (phy, mac, wifiStaNodes);Wi-Fi is configured for all our STA nodes ✨ now we gotta do the AP (access point) node. We start by changing the default Attributes of the WifiMacHelper to match what the AP needs:
mac.SetType ("ns3::ApWifiMac",
"Ssid", SsidValue (ssid));Here, the WifiMacHelper is gonna create MAC layers of the "ns3::ApWifiMac" type — that latter part says a MAC instance configured as an AP should be created. We don't touch the default setting of the "QosSupported" Attribute, so it stays false — which disables 802.11e/WMM-style QoS support at created APs.
Next lines create the single AP that shares the same set of PHY-level Attributes (and channel) as the stations:
NetDeviceContainer apDevices;
apDevices = wifi.Install (phy, mac, wifiApNode);Now, mobility models 🌈. We want the STA nodes to be mobile, wandering around inside a bounding box, and we want the AP node to stay put. The MobilityHelper makes this easy. First, instantiate a MobilityHelper object and set some Attributes controlling the "position allocator" functionality:
MobilityHelper mobility;
mobility.SetPositionAllocator ("ns3::GridPositionAllocator",
"MinX", DoubleValue (0.0),
"MinY", DoubleValue (0.0),
"DeltaX", DoubleValue (5.0),
"DeltaY", DoubleValue (10.0),
"GridWidth", UintegerValue (3),
"LayoutType", StringValue ("RowFirst"));This tells the mobility helper to use a two-dimensional grid to initially place the STA nodes. Feel free to explore the Doxygen for class ns3::GridPositionAllocator to see exactly what's going on 🦋.
Nodes are on an initial grid, but now we need to tell them how to move. We pick RandomWalk2dMobilityModel, which has nodes move in a random direction at a random speed around inside a bounding box:
mobility.SetMobilityModel ("ns3::RandomWalk2dMobilityModel",
"Bounds", RectangleValue (Rectangle (-50, 50, -50, 50)));Now we tell the MobilityHelper to install the mobility models on the STA nodes:
mobility.Install (wifiStaNodes);We want the access point to stay in a fixed position the whole simulation. We do that by setting this node's mobility model to ns3::ConstantPositionMobilityModel:
mobility.SetMobilityModel ("ns3::ConstantPositionMobilityModel");
mobility.Install (wifiApNode);So now we've got nodes, devices and channels created, mobility models picked for the Wi-Fi nodes — but no protocol stacks. Just like we've done a million times before, we use the InternetStackHelper to install those stacks:
InternetStackHelper stack;
stack.Install (csmaNodes);
stack.Install (wifiApNode);
stack.Install (wifiStaNodes);Same as in the second.cc example script, we're using Ipv4AddressHelper to hand out IP addresses to our device interfaces. First we use network 10.1.1.0 to create the two addresses our two point-to-point devices need. Then network 10.1.2.0 to assign addresses to the CSMA network, then addresses from network 10.1.3.0 to both the STA devices and the AP on the wireless network:
Ipv4AddressHelper address;
address.SetBase ("10.1.1.0", "255.255.255.0");
Ipv4InterfaceContainer p2pInterfaces;
p2pInterfaces = address.Assign (p2pDevices);
address.SetBase ("10.1.2.0", "255.255.255.0");
Ipv4InterfaceContainer csmaInterfaces;
csmaInterfaces = address.Assign (csmaDevices);
address.SetBase ("10.1.3.0", "255.255.255.0");
address.Assign (staDevices);
address.Assign (apDevices);We put the echo server on the "rightmost" node in the illustration at the start of the file. Done this before 💅:
UdpEchoServerHelper echoServer (9);
ApplicationContainer serverApps = echoServer.Install (csmaNodes.Get (nCsma));
serverApps.Start (Seconds (1.0));
serverApps.Stop (Seconds (10.0));And we put the echo client on the last STA node we created, pointing it at the server on the CSMA network. Also seen this before:
UdpEchoClientHelper echoClient (csmaInterfaces.GetAddress (nCsma), 9);
echoClient.SetAttribute ("MaxPackets", UintegerValue (1));
echoClient.SetAttribute ("Interval", TimeValue (Seconds (1.0)));
echoClient.SetAttribute ("PacketSize", UintegerValue (1024));
ApplicationContainer clientApps =
echoClient.Install (wifiStaNodes.Get (nWifi - 1));
clientApps.Start (Seconds (2.0));
clientApps.Stop (Seconds (10.0));Since we built an internetwork here, we need internetwork routing enabled, same as the second.cc example script:
Ipv4GlobalRoutingHelper::PopulateRoutingTables ();okay so here's a thing that surprises people: the simulation we just made will never "naturally" stop 😭. That's because we asked the wireless access point to generate beacons. It'll generate beacons forever, which means simulator events get scheduled into the future indefinitely — so we have to tell the simulator to stop even though it might have beacon generation events still scheduled. The line below tells the simulator to stop so we don't simulate beacons forever and fall into what's basically an endless loop:
Simulator::Stop (Seconds (10.0));We create just enough tracing to cover all three networks:
pointToPoint.EnablePcapAll ("third");
phy.EnablePcap ("third", apDevices.Get (0));
csma.EnablePcap ("third", csmaDevices.Get (0), true);Those three lines start pcap tracing on both point-to-point nodes serving as our backbone, start a promiscuous (monitor) mode trace on the Wi-Fi network, and start a promiscuous trace on the CSMA network. So we get to see all the traffic with the fewest possible trace files ✨.
Finally, we actually run the simulation, clean up, and exit the program:
Simulator::Run ();
Simulator::Destroy ();
return 0;
}To run this example you have to copy the third.cc example script into the scratch directory and build with Waf, just like with the second.cc example. From the top-level directory of the repository you'd type:
$ cp examples/tutorial/third.cc scratch/mythird.cc
$ ./waf
$ ./waf --run scratch/mythirdAgain, since the UDP echo applications are set up just like in second.cc, you'll see similar output:
Waf: Entering directory `/home/craigdo/repos/ns-3-allinone/ns-3-dev/build'
Waf: Leaving directory `/home/craigdo/repos/ns-3-allinone/ns-3-dev/build'
'build' finished successfully (0.407s)
At time 2s client sent 1024 bytes to 10.1.2.4 port 9
At time 2.01796s server received 1024 bytes from 10.1.3.3 port 49153
At time 2.01796s server sent 1024 bytes to 10.1.3.3 port 49153
At time 2.03364s client received 1024 bytes from 10.1.2.4 port 9Remember: that first message, Sent 1024 bytes to 10.1.2.4, is the UDP echo client sending a packet to the server. In this case, the client is on the wireless network (10.1.3.0). The second message, Received 1024 bytes from 10.1.3.3, is from the UDP echo server, generated when it receives the echo packet. The final message, Received 1024 bytes from 10.1.2.4, is from the echo client, showing it got its echo back from the server. 👑
If you go look in the top level directory now, you'll find four trace files from this simulation — two from node zero and two from node one:
third-0-0.pcap third-0-1.pcap third-1-0.pcap third-1-1.pcapThe file third-0-0.pcap is literally the point-to-point device on node zero — aka the left side of the "backbone" 🦋. The file third-1-0.pcap is the point-to-point device on node one — the right side of the "backbone". And third-0-1.pcap is gonna be the promiscuous (monitor mode) trace from the Wi-Fi network, while third-1-1.pcap is the promiscuous trace from the CSMA network. Can you verify this by inspecting the code? 👀
Since the echo client is chillin' on the Wi-Fi network, let's start there. Let's take a look at the promiscuous (monitor mode) trace we captured on that network.
$ tcpdump -nn -tt -r third-0-1.pcapYou should see some wifi-looking contents you haven't seen here before:
reading from file third-0-1.pcap, link-type IEEE802_11 (802.11)
0.000025 Beacon (ns-3-ssid) [6.0* 9.0 12.0 18.0 24.0 36.0 48.0 54.0 Mbit] IBSS
0.000308 Assoc Request (ns-3-ssid) [6.0 9.0 12.0 18.0 24.0 36.0 48.0 54.0 Mbit]
0.000324 Acknowledgment RA:00:00:00:00:00:08
0.000402 Assoc Response AID(0) :: Successful
0.000546 Acknowledgment RA:00:00:00:00:00:0a
0.000721 Assoc Request (ns-3-ssid) [6.0 9.0 12.0 18.0 24.0 36.0 48.0 54.0 Mbit]
0.000737 Acknowledgment RA:00:00:00:00:00:07
0.000824 Assoc Response AID(0) :: Successful
0.000968 Acknowledgment RA:00:00:00:00:00:0a
0.001134 Assoc Request (ns-3-ssid) [6.0 9.0 12.0 18.0 24.0 36.0 48.0 54.0 Mbit]
0.001150 Acknowledgment RA:00:00:00:00:00:09
0.001273 Assoc Response AID(0) :: Successful
0.001417 Acknowledgment RA:00:00:00:00:00:0a
0.102400 Beacon (ns-3-ssid) [6.0* 9.0 12.0 18.0 24.0 36.0 48.0 54.0 Mbit] IBSS
0.204800 Beacon (ns-3-ssid) [6.0* 9.0 12.0 18.0 24.0 36.0 48.0 54.0 Mbit] IBSS
0.307200 Beacon (ns-3-ssid) [6.0* 9.0 12.0 18.0 24.0 36.0 48.0 54.0 Mbit] IBSSYou can see the link type is now 802.11 like you'd expect 💅. You can probably understand what's going on and find the IP echo request and response packets in this trace. We leave it as an exercise to completely parse the trace dump, bestie — that one's on you ✨.
Now, look at the pcap file of the left side of the point-to-point link,
$ tcpdump -nn -tt -r third-0-0.pcapAgain, you should see some familiar looking contents:
reading from file third-0-0.pcap, link-type PPP (PPP)
2.008151 IP 10.1.3.3.49153 > 10.1.2.4.9: UDP, length 1024
2.026758 IP 10.1.2.4.9 > 10.1.3.3.49153: UDP, length 1024This is the echo packet going from left to right (from Wi-Fi to CSMA) and back again across the point-to-point link 🔁.
Now, look at the pcap file of the right side of the point-to-point link,
$ tcpdump -nn -tt -r third-1-0.pcapAgain, you should see some familiar looking contents:
reading from file third-1-0.pcap, link-type PPP (PPP)
2.011837 IP 10.1.3.3.49153 > 10.1.2.4.9: UDP, length 1024
2.023072 IP 10.1.2.4.9 > 10.1.3.3.49153: UDP, length 1024This is also the echo packet going from left to right (from Wi-Fi to CSMA) and back again across the point-to-point link, just with slightly different timings as you might expect ⏱️.
The echo server is on the CSMA network, so let's look at the promiscuous trace there:
$ tcpdump -nn -tt -r third-1-1.pcapYou should see some familiar looking contents:
reading from file third-1-1.pcap, link-type EN10MB (Ethernet)
2.017837 ARP, Request who-has 10.1.2.4 (ff:ff:ff:ff:ff:ff) tell 10.1.2.1, length 50
2.017861 ARP, Reply 10.1.2.4 is-at 00:00:00:00:00:06, length 50
2.017861 IP 10.1.3.3.49153 > 10.1.2.4.9: UDP, length 1024
2.022966 ARP, Request who-has 10.1.2.1 (ff:ff:ff:ff:ff:ff) tell 10.1.2.4, length 50
2.022966 ARP, Reply 10.1.2.1 is-at 00:00:00:00:00:03, length 50
2.023072 IP 10.1.2.4.9 > 10.1.3.3.49153: UDP, length 1024This should be easily understood. If you've forgotten, go back and look at the discussion in second.cc. This is the same sequence 👑.
Now, we spent a lot of time setting up mobility models for the wireless network, so it would be a total shame to finish up without even showing that the STA nodes are actually moving around during the simulation 🚶♀️✨. Let's do this by hooking into the MobilityModel course change trace source. This is just a sneak peek into the detailed tracing section which is coming up, but this seems like a very nice place to get an example in.
As mentioned in the "Tweaking ns-3" section, the ns-3 tracing system is divided into trace sources and trace sinks, and we provide functions to connect the two. We will use the mobility model predefined course change trace source to originate the trace events. We'll need to write a trace sink to connect to that source that will display some pretty information for us 💖. Despite its reputation as being difficult, it's really quite simple. Just before the main program of the scratch/mythird.cc script (i.e., just after the NS_LOG_COMPONENT_DEFINE statement), add the following function:
void
CourseChange (std::string context, Ptr<const MobilityModel> model)
{
Vector position = model->GetPosition ();
NS_LOG_UNCOND (context <<
" x = " << position.x << ", y = " << position.y);
}This code just pulls the position information from the mobility model and unconditionally logs the x and y position of the node 📍. We're going to arrange for this function to be called every time the wireless node with the echo client changes its position. We do this using the Config::Connect function. Add the following lines of code to the script just before the Simulator::Run call.
std::ostringstream oss;
oss <<
"/NodeList/" << wifiStaNodes.Get (nWifi - 1)->GetId () <<
"/$ns3::MobilityModel/CourseChange";
Config::Connect (oss.str (), MakeCallback (&CourseChange));What we do here is create a string containing the tracing namespace path of the event to which we want to connect. First, we have to figure out which node it is we want using the GetId method as described earlier. In the case of the default number of CSMA and wireless nodes, this turns out to be node seven and the tracing namespace path to the mobility model would look like,
/NodeList/7/$ns3::MobilityModel/CourseChangeBased on the discussion in the tracing section, you may infer that this trace path references the seventh node in the global NodeList. It specifies what is called an aggregated object of type ns3::MobilityModel. The dollar sign prefix implies that the MobilityModel is aggregated to node seven. The last component of the path means that we are hooking into the "CourseChange" event of that model 🔗.
We make a connection between the trace source in node seven with our trace sink by calling Config::Connect and passing this namespace path. Once this is done, every course change event on node seven will be hooked into our trace sink, which will in turn print out the new position 💫.
If you now run the simulation, you will see the course changes displayed as they happen.
'build' finished successfully (5.989s)
/NodeList/7/$ns3::MobilityModel/CourseChange x = 10, y = 0
/NodeList/7/$ns3::MobilityModel/CourseChange x = 10.3841, y = 0.923277
/NodeList/7/$ns3::MobilityModel/CourseChange x = 10.2049, y = 1.90708
/NodeList/7/$ns3::MobilityModel/CourseChange x = 10.8136, y = 1.11368
/NodeList/7/$ns3::MobilityModel/CourseChange x = 10.8452, y = 2.11318
/NodeList/7/$ns3::MobilityModel/CourseChange x = 10.9797, y = 3.10409
At time 2s client sent 1024 bytes to 10.1.2.4 port 9
At time 2.01796s server received 1024 bytes from 10.1.3.3 port 49153
At time 2.01796s server sent 1024 bytes to 10.1.3.3 port 49153
At time 2.03364s client received 1024 bytes from 10.1.2.4 port 9
/NodeList/7/$ns3::MobilityModel/CourseChange x = 11.3273, y = 4.04175
/NodeList/7/$ns3::MobilityModel/CourseChange x = 12.013, y = 4.76955
/NodeList/7/$ns3::MobilityModel/CourseChange x = 12.4317, y = 5.67771
/NodeList/7/$ns3::MobilityModel/CourseChange x = 11.4607, y = 5.91681
/NodeList/7/$ns3::MobilityModel/CourseChange x = 12.0155, y = 6.74878
/NodeList/7/$ns3::MobilityModel/CourseChange x = 13.0076, y = 6.62336
/NodeList/7/$ns3::MobilityModel/CourseChange x = 12.6285, y = 5.698
/NodeList/7/$ns3::MobilityModel/CourseChange x = 13.32, y = 4.97559
/NodeList/7/$ns3::MobilityModel/CourseChange x = 13.1134, y = 3.99715
/NodeList/7/$ns3::MobilityModel/CourseChange x = 13.8359, y = 4.68851
/NodeList/7/$ns3::MobilityModel/CourseChange x = 13.5953, y = 3.71789
/NodeList/7/$ns3::MobilityModel/CourseChange x = 12.7595, y = 4.26688
/NodeList/7/$ns3::MobilityModel/CourseChange x = 11.7629, y = 4.34913
/NodeList/7/$ns3::MobilityModel/CourseChange x = 11.2292, y = 5.19485
/NodeList/7/$ns3::MobilityModel/CourseChange x = 10.2344, y = 5.09394
/NodeList/7/$ns3::MobilityModel/CourseChange x = 9.3601, y = 4.60846
/NodeList/7/$ns3::MobilityModel/CourseChange x = 8.40025, y = 4.32795
/NodeList/7/$ns3::MobilityModel/CourseChange x = 9.14292, y = 4.99761
/NodeList/7/$ns3::MobilityModel/CourseChange x = 9.08299, y = 5.99581
/NodeList/7/$ns3::MobilityModel/CourseChange x = 8.26068, y = 5.42677
/NodeList/7/$ns3::MobilityModel/CourseChange x = 8.35917, y = 6.42191
/NodeList/7/$ns3::MobilityModel/CourseChange x = 7.66805, y = 7.14466
/NodeList/7/$ns3::MobilityModel/CourseChange x = 6.71414, y = 6.84456
/NodeList/7/$ns3::MobilityModel/CourseChange x = 6.42489, y = 7.80181🎀 Queues in ns-3 🎀
The selection of queueing disciplines in ns-3 can have a huge impact on performance, and it is important for users to understand what is installed by default and how to change the defaults and observe the performance 👀🔥.
Architecturally, ns-3 separates the device layer from the IP layers or traffic control layers of an Internet host. Since recent releases of ns-3, outgoing packets traverse two queueing layers before reaching the channel object. The first queueing layer encountered is what is called the 'traffic control layer' in ns-3; here, active queue management (RFC7567) and prioritization due to quality-of-service (QoS) takes place in a device-independent manner through the use of queueing disciplines. The second queueing layer is typically found in the NetDevice objects. Different devices (e.g. LTE, Wi-Fi) have different implementations of these queues. This two-layer approach mirrors what is found in practice, (software queues providing prioritization, and hardware queues specific to a link type). In practice, it may be even more complex than this. For instance, address resolution protocols have a small queue. Wi-Fi in Linux has four layers of queueing (- https://lwn.net/Articles/705884/).
The traffic control layer is only actually doing its job if the NetDevice yells at it the moment the device queue is full — like, that's the signal it needs to stop yeeting packets at the NetDevice. No signal? Then the backlog of the queueing disciplines is always null and they're just sitting there being useless. 💀
Currently, flow control — i.e. that whole "ability to notify the traffic control layer" thing — is supported by these NetDevices, which use Queue objects (or objects of Queue subclasses) to store their packets:
- Point-To-Point
- Csma
- Wi-Fi
- SimpleNetDevice
The performance of queueing disciplines is super impacted by how big the queues used by the NetDevices are. Currently, queues by default in ns-3 are not autotuned for the configured link properties (bandwidth, delay), and are typically the simplest variants (e.g. FIFO scheduling with drop-tail behavior). ngl, kinda basic
But bestie — the size of the queues can be dynamically adjusted by enabling BQL (Byte Queue Limits), the algorithm implemented in the Linux kernel to adjust the size of the device queues to fight bufferbloat while avoiding starvation. Currently, BQL is supported by the NetDevices that support flow control. ✨
An analysis of the impact of the size of the device queues on the effectiveness of the queueing disciplines, conducted by means of ns-3 simulations and real experiments, is reported in:
P. Imputato and S. Avallone. An analysis of the impact of network device buffers on packet schedulers through experiments and simulations. Simulation Modelling Practice and Theory, 80(Supplement C):1–18, January 2018. DOI: 10.1016/j.simpat.2017.09.008
✨ Available queueing models in ns-3 ✨
At the traffic-control layer, these are the options, and yes we're listing every single one:
PFifoFastQueueDisc: The default maximum size is 1000 packetsFifoQueueDisc: The default maximum size is 1000 packetsRedQueueDisc: The default maximum size is 25 packetsCoDelQueueDisc: The default maximum size is 1500 kilobytesFqCoDelQueueDisc: The default maximum size is 10240 packetsPieQueueDisc: The default maximum size is 25 packetsMqQueueDisc: This queue disc has no limits on its capacityTbfQueueDisc: The default maximum size is 1000 packets
By default, a pfifo_fast queueing discipline is installed on a NetDevice when an IPv4 or IPv6 address is assigned to an interface associated with the NetDevice — unless a queueing discipline has been already installed on the NetDevice. (It knows when you've already decorated, it's not gonna double up 💅)
At the device layer, there are device specific queues:
PointToPointNetDevice: The default configuration (as set by the helper) is to install a DropTail queue of default size (100 packets)CsmaNetDevice: The default configuration (as set by the helper) is to install a DropTail queue of default size (100 packets)WiFiNetDevice: The default configuration is to install a DropTail queue of default size (100 packets) for non-QoS stations and four DropTail queues of default size (100 packets) for QoS stationsSimpleNetDevice: The default configuration is to install a DropTail queue of default size (100 packets)LTENetDevice: Queueing occurs at the RLC layer (RLC UM default buffer is 10 * 1024 bytes, RLC AM does not have a buffer limit).UanNetDevice: There is a default 10 packet queue at the MAC layer
🦋 Changing from the defaults 🦋
- The type of queue used by a NetDevice can be usually modified through the device helper:
NodeContainer nodes;
nodes.Create (2);
PointToPointHelper p2p;
p2p.SetQueue ("ns3::DropTailQueue", "MaxSize", StringValue ("50p"));
NetDeviceContainer devices = p2p.Install (nodes);- The type of queue disc installed on a NetDevice can be modified through the traffic control helper:
InternetStackHelper stack;
stack.Install (nodes);
TrafficControlHelper tch;
tch.SetRootQueueDisc ("ns3::CoDelQueueDisc", "MaxSize", StringValue ("1000p"));
tch.Install (devices);- BQL can be enabled on a device that supports it through the traffic control helper:
InternetStackHelper stack;
stack.Install (nodes);
TrafficControlHelper tch;
tch.SetRootQueueDisc ("ns3::CoDelQueueDisc", "MaxSize", StringValue ("1000p"));
tch.SetQueueLimits ("ns3::DynamicQueueLimits", "HoldTime", StringValue ("4ms"));
tch.Install (devices);🌈 TRACING 🌈
💜 Background 💜
So, like it says over in UsingTracingSystem, the whole entire point of running an ns-3 simulation is to generate output you can actually study. And you've got two basic strategies for getting output out of ns-3: use the generic pre-defined bulk output mechanisms and then parse their content to extract the interesting stuff, or somehow build your own output mechanism that conveys exactly (and maybe only) the information you want. 👑
Using pre-defined bulk output mechanisms has the advantage of not requiring any changes to ns-3, but it may require writing scripts to parse and filter for data of interest. Often, PCAP or NS_LOG output messages are gathered during simulation runs and separately run through scripts that use grep, sed or awk to parse the messages and reduce and transform the data to a manageable form. Programs must be written to do the transformation, so this does not come for free. 🙈
NS_LOG output is not considered part of the ns-3 API, and can change without warning between releases. In addition, NS_LOG output is only available in debug builds, so relying on it imposes a performance penalty. And of course, if the information of interest doesn't exist in any of the pre-defined output mechanisms, this whole approach just... fails. 💀
If you need to add some tidbit of information to the pre-defined bulk mechanisms, this can certainly be done; and if you use one of the ns-3 mechanisms, you may get your code added as a contribution. ✨ (main character energy, honestly)
ns-3 provides another mechanism, called Tracing, that avoids some of the problems inherent in the bulk output mechanisms. It has several important advantages. First, you can reduce the amount of data you have to manage by only tracing the events of interest to you (for large simulations, dumping everything to disk for post-processing can create I/O bottlenecks). Second, if you use this method, you can control the format of the output directly so you avoid the postprocessing step with sed, awk, perl or python scripts. If you desire, your output can be formatted directly into a form acceptable by gnuplot, for example (see also GnuplotHelper). You can add hooks in the core which can then be accessed by other users, but which will produce no information unless explicitly asked to do so.
For these reasons, we believe that the ns-3 tracing system is the best way to get information out of a simulation and is also therefore one of the most important mechanisms to understand in ns-3. 🫶
🎀 Blunt Instruments 🎀
There are many ways to get information out of a program. The most straightforward way is to just print the information directly to the standard output, as in:
#include <iostream>
...
void
SomeFunction (void)
{
uint32_t x = SOME_INTERESTING_VALUE;
...
std::cout << "The value of x is " << x << std::endl;
...
}Nobody is going to prevent you from going deep into the core of ns-3 and adding print statements. It's insanely easy to do and, after all, you have complete control of your own ns-3 branch. This will probably not turn out to be very satisfactory in the long term, though. 😭
As the number of print statements increases in your programs, the task of dealing with the large number of outputs will become more and more complicated. Eventually, you may feel the need to control what information is being printed in some way, perhaps by turning on and off certain categories of prints, or increasing or decreasing the amount of information you want. If you continue down this path you may discover that you have re-implemented the NS_LOG mechanism (see UsingLogging). In order to avoid that, one of the first things you might consider is using NS_LOG itself. 💅
We mentioned above that one way to get information out of ns-3 is to parse existing NS_LOG output for interesting information. If you discover that some tidbit of information you need is not present in existing log output, you could edit the core of ns-3 and simply add your interesting information to the output stream. Now, this is certainly better than adding your own print statements since it follows ns-3 coding conventions and could potentially be useful to other people as a patch to the existing core.
Let's pick a random example. If you wanted to add more logging to the ns-3 TCP socket (tcp-socket-base.cc) you could just add a new message down in the implementation. Notice that in TcpSocketBase::ProcessEstablished () there is no log message for the reception of a SYN+ACK in ESTABLISHED state. You could simply add one, changing the code. Here is the original:
/* Received a packet upon ESTABLISHED state. This function is mimicking the
role of tcp_rcv_established() in tcp_input.c in Linux kernel. */
void
TcpSocketBase::ProcessEstablished (Ptr<Packet> packet, const TcpHeader& tcpHeader)
{
NS_LOG_FUNCTION (this << tcpHeader);
...
else if (tcpflags == (TcpHeader::SYN | TcpHeader::ACK))
{ // No action for received SYN+ACK, it is probably a duplicated packet
}
...To log the SYN+ACK case, you can add a new NS_LOG_LOGIC in the if statement body:
/* Received a packet upon ESTABLISHED state. This function is mimicking the
role of tcp_rcv_established() in tcp_input.c in Linux kernel. */
void
TcpSocketBase::ProcessEstablished (Ptr<Packet> packet, const TcpHeader& tcpHeader)
{
NS_LOG_FUNCTION (this << tcpHeader);
...
else if (tcpflags == (TcpHeader::SYN | TcpHeader::ACK))
{ // No action for received SYN+ACK, it is probably a duplicated packet
NS_LOG_LOGIC ("TcpSocketBase " << this << " ignoring SYN+ACK");
}
...This may seem fairly simple and satisfying at first glance, but something to consider is that you will be writing code to add NS_LOG statements and you will also have to write code (as in grep, sed or awk scripts) to parse the log output in order to isolate your information. This is because even though you have some control over what is output by the logging system, you only have control down to the log component level, which is typically an entire source code file. 😭✨
Like, if you're adding code to an existing module, you also gotta live with whatever output every other developer thought was interesting. So you'll probably find that to get that one tiny crumb of info you actually need, you have to wade through huge amounts of extraneous messages that are of no interest to you at all. You may literally be forced to save giant log files to disk and process them down to a few lines every single time you want to do anything. 😭
And since there are no guarantees in ns-3 about the stability of NS_LOG output, you may also discover that pieces of log output you depend on just… disappear or change between releases. 🙈 If you depend on the structure of the output, you may find other messages being added or deleted, which can totally wreck your parsing code.
Finally: NS_LOG output is only available in debug builds. You can't get log output from optimized builds, which run about twice as fast. So relying on NS_LOG imposes a performance penalty. 💀
For these reasons, we consider prints to std::cout and NS_LOG messages to be quick and dirty ways to get more information out of ns-3 — not suitable for serious work.
What we actually want is a stable facility, using stable APIs, that lets you reach into the core system and only get the information required. Ideally you could do it without having to change and recompile the core system. And even better? A system that notified user code when an item of interest changed, or an interesting event happened, so the user doesn't have to actively poke around in the system looking for things.
The ns-3 tracing system is designed to work along exactly those lines, and it's well-integrated with the Attribute and Config subsystems, which allows for relatively simple use scenarios. 👑
✨ Overview ✨
The ns-3 tracing system is built on the concepts of independent tracing sources and tracing sinks, plus a uniform mechanism for connecting sources to sinks.
Trace sources are entities that can signal events that happen in a simulation and provide access to interesting underlying data. For example, a trace source could indicate when a packet is received by a net device and provide access to the packet contents for interested trace sinks. A trace source might also indicate when an interesting state change happens in a model. For example, the congestion window of a TCP model is a prime candidate for a trace source. Every time the congestion window changes, connected trace sinks are notified with the old and new value. 🎀
Trace sources are not useful by themselves; they must be connected to other pieces of code that actually do something useful with the information provided by the source. The entities that consume trace information are called trace sinks. Trace sources are generators of data, and trace sinks are consumers. This explicit division lets large numbers of trace sources be scattered around the system in places which model authors believe might be useful. Inserting trace sources introduces a very small execution overhead.
There can be zero or more consumers of trace events generated by a trace source. You can think of a trace source as a kind of point-to-multipoint information link. Your code looking for trace events from a particular piece of core code could happily coexist with other code doing something entirely different from the same information. 🦋
Unless a user connects a trace sink to one of these sources, nothing is output. By using the tracing system, both you and other people hooked to the same trace source are getting exactly what they want, and only what they want, out of the system. Neither of you is impacting any other user by changing what information is output by the system. And if you happen to add a trace source, your work as a good open-source citizen may allow other users to provide new utilities that are perhaps very useful overall — without making any changes to the ns-3 core. 💅
🌸 Simple Example 🌸
Let's take a few minutes and walk through a simple tracing example. We're gonna need a little background on Callbacks to understand what's happening in the example, so we have to take a small detour right away. 🚗
🧚 Callbacks 🧚
The goal of the Callback system in ns-3 is to allow one piece of code to call a function (or method in C++) without any specific inter-module dependency. This ultimately means you need some kind of indirection — you treat the address of the called function as a variable. This variable is called a pointer-to-function variable. The relationship between function and pointer-to-function is really no different than that of object and pointer-to-object. fr fr.
In C, the canonical example of a pointer-to-function is a pointer-to-function-returning-integer (PFI). For a PFI taking one int parameter, this could be declared like,
int (*pfi)(int arg) = 0;(But read the C++-FAQ Section 33 before writing code like this!) What you get from this is a variable named simply pfi that is initialized to the value 0. If you want to initialize this pointer to something meaningful, you need to have a function with a matching signature. In this case, you could provide a function that looks like:
int MyFunction (int arg) {}If you have this target, you can initialize the variable to point to your function:
pfi = MyFunction;You can then call MyFunction indirectly using the more suggestive form of the call:
int result = (*pfi) (1234);This is suggestive since it looks like you're dereferencing the function pointer just like you would dereference any pointer. Typically, however, people take advantage of the fact that the compiler knows what's going on and will just use a shorter form:
int result = pfi (1234);This looks like you're calling a function named pfi, but the compiler is smart enough to know to call through the variable pfi indirectly to the function MyFunction. ✨
Conceptually, this is almost exactly how the tracing system works. Basically, a trace sink is a callback. When a trace sink expresses interest in receiving trace events, it adds itself as a Callback to a list of Callbacks internally held by the trace source. When an interesting event happens, the trace source invokes its operator(...) providing zero or more arguments. The operator(...) eventually wanders down into the system and does something remarkably like the indirect call you just saw, providing zero or more parameters — just as the call to pfi above passed one parameter to the target function MyFunction.
The important difference that the tracing system adds is that for each trace source there is an internal list of Callbacks. Instead of just making one indirect call, a trace source may invoke multiple Callbacks. When a trace sink expresses interest in notifications from a trace source, it basically just arranges to add its own function to the callback list. 🌈
If you're interested in more details about how this is actually arranged in ns-3, feel free to peruse the Callback section of the ns-3 Manual.
👑 Walkthrough: fourth.cc 👑
We've provided some code to implement what is really the simplest example of tracing that can be assembled. You can find this code in the tutorial directory as fourth.cc. Let's walk through it:
/* -*- Mode:C++; c-file-style:"gnu"; indent-tabs-mode:nil; -*- */
/*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License version 2 as
* published by the Free Software Foundation;
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
*/
#include "ns3/object.h"
#include "ns3/uinteger.h"
#include "ns3/traced-value.h"
#include "ns3/trace-source-accessor.h"
#include <iostream>
using namespace ns3;Most of this code should be quite familiar to you. As mentioned above, the trace system makes heavy use of the Object and Attribute systems, so you'll need to include them. The first two includes above bring in the declarations for those systems explicitly. You could use the core module header to get everything at once, but we do the includes explicitly here to illustrate how simple this all really is. 🫶
The file, traced-value.h brings in the required declarations for tracing of data that obeys value semantics. In general, value semantics just means that you can pass the object itself around, rather than passing the address of the object. What this all really means is that you'll be able to trace all changes made to a TracedValue in a really simple way. 💖
Since the tracing system is integrated with Attributes, and Attributes work with Objects, there must be an ns-3 Object for the trace source to live in. The next code snippet declares and defines a simple Object we can work with.
class MyObject : public Object
{
public:
static TypeId GetTypeId (void)
{
static TypeId tid = TypeId ("MyObject")
.SetParent (Object::GetTypeId ())
.SetGroupName ("MyGroup")
.AddConstructor<MyObject> ()
.AddTraceSource ("MyInteger",
"An integer value to trace.",
MakeTraceSourceAccessor (&MyObject::m_myInt),
"ns3::TracedValueCallback::Int32")
;
return tid;
}
MyObject () {}
TracedValue<int32_t> m_myInt;
};The two important lines of code, above, with respect to tracing are the .AddTraceSource and the TracedValue declaration of m_myInt.
The .AddTraceSource provides the "hooks" used for connecting the trace source to the outside world through the Config system. The first argument is a name for this trace source, which makes it visible in the Config system. The second argument is a help string. Now look at the third argument — in fact, focus on the argument of the third argument: &MyObject::m_myInt. This is the TracedValue which is being added to the class; it is always a class data member. (The final argument is the name of a typedef for the TracedValue type, as a string. This is used to generate documentation for the correct Callback function signature, which is useful especially for more general types of Callbacks.) ⭐
The TracedValue<> declaration is literally the ✨infrastructure✨ that runs the whole callback situation. Anytime the underlying value changes, the TracedValue mechanism hands you both the old value and the new value — in this case an int32_t. So the trace sink function traceSink for this TracedValue needs the signature:
void (* traceSink)(int32_t oldValue, int32_t newValue);Every single trace sink that hooks this trace source must have that signature, no exceptions bestie. (We'll get into how you figure out the required callback signature in other cases down below. 🫶)
And sure enough, if you keep scrolling through fourth.cc you see:
void
IntTrace (int32_t oldValue, int32_t newValue)
{
std::cout << "Traced " << oldValue << " to " << newValue << std::endl;
}That's the definition of a matching trace sink. It lines up directly with the callback function signature. Once it's connected, this function gets called every time the TracedValue changes. 💅
So at this point we've met the trace source and the trace sink. What's left is the code that connects the source to the sink — and that happens in main:
int
main (int argc, char *argv[])
{
Ptr<MyObject> myObject = CreateObject<MyObject> ();
myObject->TraceConnectWithoutContext ("MyInteger", MakeCallback(&IntTrace));
myObject->m_myInt = 1234;
}First we create the MyObject instance that the trace source lives inside. 🎀
Next up: TraceConnectWithoutContext, which forms the connection between the trace source and the trace sink. The first argument is just the trace source name "MyInteger" that we saw above. Now notice the MakeCallback template function — this one does the magic required to create the underlying ns-3 Callback object and associate it with the function IntTrace. TraceConnect makes the association between your provided function and the overloaded operator() in the traced variable referred to by the "MyInteger" Attribute. Once that association is made, the trace source will "fire" your provided callback function. 🔥
The code making all of this happen is, of course, non-trivial, but the essence is: you're arranging for something that looks just like the pfi() example above to be called by the trace source. The declaration of TracedValue<int32_t> m_myInt; inside the Object itself performs the magic needed to provide the overloaded assignment operators that will use the operator() to actually invoke the Callback with the desired parameters. The .AddTraceSource performs the magic to connect the Callback to the Config system, and TraceConnectWithoutContext performs the magic to connect your function to the trace source — which is specified by Attribute name. ✨
Let's ignore the bit about context for now. 🙈
Finally, the line assigning a value to m_myInt:
myObject->m_myInt = 1234;should be interpreted as an invocation of operator= on the member variable m_myInt with the integer 1234 passed as a parameter.
Since m_myInt is a TracedValue, this operator is defined to execute a callback that returns void and takes two integer values as parameters — an old value and a new value for the integer in question. That is exactly the function signature for the callback function we provided — IntTrace. 💯
To summarize, a trace source is, in essence, a variable that holds a list of callbacks. A trace sink is a function used as the target of a callback. The Attribute and object type information systems are used to provide a way to connect trace sources to trace sinks. The act of "hitting" a trace source is executing an operator on the trace source which fires callbacks. This results in the trace sink callbacks who registered interest in the source being called with the parameters provided by the source. Periodt. 💅
If you now build and run this example,
$ ./waf --run fourthyou'll see the output from the IntTrace function execute as soon as the trace source is hit:
Traced 0 to 1234When we executed the code myObject->m_myInt = 1234;, the trace source fired and automatically provided the before and after values to the trace sink. The function IntTrace then printed this to the standard output. 🦋
Connect with Config 🔌
The TraceConnectWithoutContext call shown above in the simple example is actually very rarely used in the system. More typically, the Config subsystem is used to select a trace source in the system using what's called a Config path. We saw an example of this in the previous section where we hooked the "CourseChange" event while we were experimenting with third.cc. 👑
Recall that we defined a trace sink to print course change information from the mobility models of our simulation. It should now be so much clearer what this function is doing:
void
CourseChange (std::string context, Ptr<const MobilityModel> model)
{
Vector position = model->GetPosition ();
NS_LOG_UNCOND (context <<
" x = " << position.x << ", y = " << position.y);
}When we connected the "CourseChange" trace source to the above trace sink, we used a Config path to specify the source when we arranged a connection between the pre-defined trace source and the new trace sink:
std::ostringstream oss;
oss << "/NodeList/"
<< wifiStaNodes.Get (nWifi - 1)->GetId ()
<< "/$ns3::MobilityModel/CourseChange";
Config::Connect (oss.str (), MakeCallback (&CourseChange));Let's try and make some sense of what is sometimes considered relatively mysterious code. For the purposes of discussion, assume that the Node number returned by the GetId() is "7". In this case, the path above turns out to be
"/NodeList/7/$ns3::MobilityModel/CourseChange"The last segment of a config path must be an Attribute of an Object. In fact, if you had a pointer to the Object that has the "CourseChange" Attribute handy, you could write this just like we did in the previous example. You know by now that we typically store pointers to our Nodes in a NodeContainer. In the third.cc example, the Nodes of interest are stored in the wifiStaNodes NodeContainer. In fact, while putting the path together, we used this container to get a Ptr<Node> which we used to call GetId(). We could have used this Ptr<Node> to call a Connect method directly:
Ptr<Object> theObject = wifiStaNodes.Get (nWifi - 1);
theObject->TraceConnectWithoutContext ("CourseChange", MakeCallback (&CourseChange));In the third.cc example, we actually wanted an additional "context" to be delivered along with the Callback parameters (which will be explained below), so we could actually use the following equivalent code:
Ptr<Object> theObject = wifiStaNodes.Get (nWifi - 1);
theObject->TraceConnect ("CourseChange", MakeCallback (&CourseChange));Turns out the internal code for Config::ConnectWithoutContext and Config::Connect actually find a Ptr<Object> and call the appropriate TraceConnect method at the lowest level. 🧚
The Config functions take a path that represents a chain of Object pointers. Each segment of a path corresponds to an Object Attribute. The last segment is the Attribute of interest, and prior segments must be typed to contain or find Objects. The Config code parses and "walks" this path until it gets to the final segment of the path. It then interprets the last segment as an Attribute on the last Object it found while walking the path. The Config functions then call the appropriate TraceConnect or TraceConnectWithoutContext method on the final Object. Let's see what happens in a bit more detail when the above path is walked. 🚶♀️
The leading / character in the path refers to a so-called namespace. One of the predefined namespaces in the config system is "NodeList" which is a list of all of the nodes in the simulation. Items in the list are referred to by indices into the list, so "/NodeList/7" refers to the eighth Node in the list of nodes created during the simulation (recall indices start at 0). This reference is actually a Ptr<Node>` and so is a subclass of an ns3::Object. 🌈
As described in the Object Model section of the ns-3 Manual, we make widespread use of object aggregation. This allows us to form an association between different Objects without building a complicated inheritance tree or predeciding what objects will be part of a Node. Each Object in an Aggregation can be reached from the other Objects.
In our example the next path segment being walked begins with the $ character. This indicates to the config system that the segment is the name of an object type, so a GetObject call should be made looking for that type. Turns out the MobilityHelper used in third.cc arranges to Aggregate, or associate, a mobility model to each of the wireless Nodes. When you add the $ you're asking for another Object that has presumably been previously aggregated. You can think of this as switching pointers from the original Ptr<Node> as specified by "/NodeList/7" to its associated mobility model — which is of type ns3::MobilityModel. If you're familiar with GetObject, we've asked the system to do the following:
Ptr<MobilityModel> mobilityModel = node->GetObject<MobilityModel> ()We're now at the last Object in the path, so we turn our attention to the Attributes of that Object. The MobilityModel class defines an Attribute called "CourseChange". You can see this by looking at the source code in src/mobility/model/mobility-model.cc and searching for "CourseChange" in your favorite editor. You should find
.AddTraceSource ("CourseChange",
"The value of the position and/or velocity vector changed",
MakeTraceSourceAccessor (&MobilityModel::m_courseChangeTrace),
"ns3::MobilityModel::CourseChangeCallback")which should look very familiar at this point. 👀
If you look for the corresponding declaration of the underlying traced variable in mobility-model.h you'll find
TracedCallback<Ptr<const MobilityModel> > m_courseChangeTrace;The type declaration TracedCallback identifies m_courseChangeTrace as a special list of Callbacks that can be hooked using the Config functions described above. The typedef for the callback function signature is also defined in the header file:
typedef void (* CourseChangeCallback)(Ptr<const MobilityModel> * model);The MobilityModel class is designed to be a base class providing a common interface for all of the specific subclasses. If you search down to the end of the file, you'll see a method defined called NotifyCourseChange():
void
MobilityModel::NotifyCourseChange (void) const
{
m_courseChangeTrace(this);
}Finding Sources
So the very first question every baby tracing user asks is: "okay so I know there are trace sources in the simulation core somewhere, but how do I actually find out which ones are available to me??" 💀
Then the follow-ups roll in, in order:
- "okay, found a trace source, but how do I figure out the Config path to use when I connect to it?"
- "okay, found the trace source AND the Config path, but how do I know what the return type and formal arguments of my callback function need to be?"
- "okay, I typed all of that in and got this incredibly bizarre error message, what in the world does it even mean??"
We'll take them one at a time, bestie. No rush. 🫶
Available Sources
Okay, I know that there must be trace sources in the simulation core, but how do I find out what trace sources are available to me?
Answer number one lives in the ns-3 API documentation. Go to the project web site, ns-3 project, and you'll spot a link to "Documentation" in the navigation bar. Click that, and you land on the documentation page. From there, a "Latest Release" link takes you to the docs for the newest stable release of ns-3. Hit the "API Documentation" link and boom — you're on the ns-3 API documentation page. ✨
In the sidebar you should see a hierarchy that begins:
- ns-3
- ns-3 Documentation
- All TraceSources
- All Attributes
- All GlobalValues
The one we care about here is "All TraceSources". Go on, click it. You will see — perhaps not too surprisingly 💅 — a list of all the trace sources available in ns-3.
As an example, scroll down to ns3::MobilityModel. You'll find an entry for:
CourseChange: The value of the position and/or velocity vector changedYou should recognize this as the trace source we used in the third.cc example. Perusing this list is genuinely helpful, no cap. 👑
Config Paths
Okay, I found a trace source, how do I figure out the Config path to use when I connect to it?
If you already know which object you're interested in, the "Detailed Description" section for that class will list all available trace sources. So, starting from the list of "All TraceSources," click the ns3::MobilityModel link, and you'll land on the documentation for the MobilityModel class. Almost at the top of the page there's a one-line brief description of the class, ending in a "More…" link. Click that to skip the API summary and jump straight to the "Detailed Description" of the class. At the end of the description you'll find (up to) three lists:
- Config Paths: a list of typical Config paths for this class.
- Attributes: a list of all attributes supplied by this class.
- TraceSources: a list of all TraceSources available from this class.
First up: the Config paths. 🌸
Let's say you just found the "CourseChange" trace source in the "All TraceSources" list and now you want to know how to connect to it. You know you're using (again, from the third.cc example) an ns3::RandomWalk2dMobilityModel. So either click the class name in the "All TraceSources" list, or hunt down ns3::RandomWalk2dMobilityModel in the "Class List". Either way, you should now be staring at the "ns3::RandomWalk2dMobilityModel Class Reference" page.
Now scroll down to the "Detailed Description" section — after the summary list of class methods and attributes (or just click the "More…" link at the end of the class brief description at the top of the page) — and you'll see the overall documentation for the class. Keep scrolling until you hit the "Config Paths" list:
Config Paths
ns3::RandomWalk2dMobilityModel is accessible through the following
paths with Config::Set and Config::Connect:
• "/NodeList/[i]/$ns3::MobilityModel/$ns3::RandomWalk2dMobilityModel"The documentation is telling you how to get to the RandomWalk2dMobilityModel Object. Now compare that string with the string we actually used in the example code:
"/NodeList/7/$ns3::MobilityModel"The difference comes from the fact that two GetObject calls are implied in the string found in the documentation. The first one, for $ns3::MobilityModel, will query the aggregation for the base class. The second implied GetObject call, for $ns3::RandomWalk2dMobilityModel, is used to cast the base class to the concrete implementation class. The documentation shows you both of those operations. And it turns out the actual trace source you're looking for is found in the base class. 🙈
Look further down in the "Detailed Description" section for the list of trace sources. You will find:
No TraceSources are defined for this type.
TraceSources defined in parent class ``ns3::MobilityModel``
• CourseChange: The value of the position and/or velocity vector
changed.
Callback signature: ns3::MobilityModel::CourseChangeCallbackThis is exactly the tea you needed. ☕ The trace source of interest lives in ns3::MobilityModel (which, like, you knew anyway). The interesting bit this piece of API documentation tells you is that you don't need that extra cast in the config path above to get to the concrete class, because the trace source is actually in the base class. So the additional GetObject is not required and you simply use the path:
"/NodeList/[i]/$ns3::MobilityModel"which perfectly matches the example path:
"/NodeList/7/$ns3::MobilityModel"As an aside — another way to find the Config path is to grep around in the ns-3 codebase for someone who already figured it out. You should always try to copy someone else's working code before you start writing your own. Try something like:
$ find . -name '*.cc' | xargs grep CourseChange | grep Connectand you may find your answer, along with working code. For example, in this case, src/mobility/examples/main-random-topology.cc has something just waiting for you to use:
Config::Connect ("/NodeList/*/$ns3::MobilityModel/CourseChange",
MakeCallback (&CourseChange));We'll come back to this example in a sec. 🎀
Callback Signatures
Okay, I found a trace source and the Config path, how do I figure out what the return type and formal arguments of my callback function need to be?
Easiest way: go look at the callback signature typedef. It's given in the "Callback signature" of the trace source in the "Detailed Description" for the class, exactly as shown above.
Repeating the "CourseChange" trace source entry from ns3::RandomWalk2dMobilityModel, we have:
• CourseChange: The value of the position and/or velocity vector
changed.
Callback signature: ns3::MobilityModel::CourseChangeCallbackThe callback signature is given as a link to the relevant typedef, where we find:
typedef void (* CourseChangeCallback)(std::string context, Ptr<const
MobilityModel> * model);
TracedCallback signature for course change notifications.
If the callback is connected using ConnectWithoutContext omit the
context argument from the signature.
Parameters:
[in] context The context string supplied by the Trace source.
[in] model The MobilityModel which is changing course.Same as before — to see this in use, grep around in the ns-3 codebase for an example. The example above, from src/mobility/examples/main-random-topology.cc, connects the "CourseChange" trace source to the CourseChange function in the same file:
static void
CourseChange (std::string context, Ptr<const MobilityModel> model)
{
...
}Notice that this function:
- Takes a "context" string argument, which we'll describe in a minute. (If the callback is connected using the
ConnectWithoutContextfunction, the context argument will be omitted.) - Has the
MobilityModelsupplied as the last argument (or the only argument ifConnectWithoutContextis used). - Returns
void.
If, by chance, the callback signature hasn't been documented, and there are no examples to work from, figuring out the right callback function signature can be, well… challenging to actually work out from the source code. 😭
Before we go on a whole walkthrough of the code, I'll be kind and just tell you a simple way to figure this out: the return value of your callback will always be void. The formal parameter list for a TracedCallback can be found from the template parameter list in the declaration. Recall that for our current example, this is in mobility-model.h, where we have previously found:
TracedCallback<Ptr<const MobilityModel> > m_courseChangeTrace;There is a one-to-one correspondence between the template parameter list in the declaration and the formal arguments of the callback function. Here, there is one template parameter, which is a Ptr<const MobilityModel>. This tells you that you need a function that returns void and takes a Ptr<const MobilityModel>. For example:
void
CourseChange (Ptr<const MobilityModel> model)
{
...
}That's literally all you need if you're gonna Config::ConnectWithoutContext. But if you want a context?? Then you need Config::Connect and you gotta use a Callback function that takes a string context, then the template arguments:
void
CourseChange (std::string context, Ptr<const MobilityModel> model)
{
...
}And if you want to make sure your CourseChangeCallback function is only visible in your local file, bestie, you can just add the static keyword and come up with:
static void
CourseChange (std::string path, Ptr<const MobilityModel> model)
{
...
}which is exactly what we used in the third.cc example. 💅
🔥 Implementation 🔥
ok so this section is entirely optional. It's gonna be a bumpy ride, especially for those of you unfamiliar with the details of templates. BUT — if you get through this, you'll have a very good handle on a lot of the ns-3 low level idioms. 🫶
So, again, let's figure out what signature of callback function is required for the "CourseChange" trace source. Yeah it's gonna be painful 😭 but you only need to do this once. After you get through this, you'll be able to just look at a TracedCallback and understand it. periodt.
First thing we need to look at is the declaration of the trace source. Recall that this is in mobility-model.h, where we have previously found:
TracedCallback<Ptr<const MobilityModel> > m_courseChangeTrace;This declaration is for a template. The template parameter is inside the angle-brackets, so what we're really interested in is finding out what that TracedCallback<> actually is. If you have absolutely no idea where this might be found — grep is your friend. 💖
We're probably gonna be interested in some kind of declaration in the ns-3 source, so first change into the src directory. Then, we know this declaration has to be in some kind of header file, so just grep for it using:
$ find . -name '*.h' | xargs grep TracedCallbackYou'll see 303 lines fly by (I piped this through wc to see how bad it was 💀). Although that may seem like a lot, that's not really a lot. Just pipe the output through more and start scanning through it. On the first page, you'll see some very suspiciously template-looking stuff.
TracedCallback<T1,T2,T3,T4,T5,T6,T7,T8>::TracedCallback ()
TracedCallback<T1,T2,T3,T4,T5,T6,T7,T8>::ConnectWithoutContext (c ...
TracedCallback<T1,T2,T3,T4,T5,T6,T7,T8>::Connect (const CallbackB ...
TracedCallback<T1,T2,T3,T4,T5,T6,T7,T8>::DisconnectWithoutContext ...
TracedCallback<T1,T2,T3,T4,T5,T6,T7,T8>::Disconnect (const Callba ...
TracedCallback<T1,T2,T3,T4,T5,T6,T7,T8>::operator() (void) const ...
TracedCallback<T1,T2,T3,T4,T5,T6,T7,T8>::operator() (T1 a1) const ...
TracedCallback<T1,T2,T3,T4,T5,T6,T7,T8>::operator() (T1 a1, T2 a2 ...
TracedCallback<T1,T2,T3,T4,T5,T6,T7,T8>::operator() (T1 a1, T2 a2 ...
TracedCallback<T1,T2,T3,T4,T5,T6,T7,T8>::operator() (T1 a1, T2 a2 ...
TracedCallback<T1,T2,T3,T4,T5,T6,T7,T8>::operator() (T1 a1, T2 a2 ...
TracedCallback<T1,T2,T3,T4,T5,T6,T7,T8>::operator() (T1 a1, T2 a2 ...
TracedCallback<T1,T2,T3,T4,T5,T6,T7,T8>::operator() (T1 a1, T2 a2 ...Turns out that all of this comes from the header file traced-callback.h which sounds very promising. You can then take a look at mobility-model.h and see that there's a line which confirms this hunch:
#include "ns3/traced-callback.h"Of course, you could have gone at this from the other direction and started by looking at the includes in mobility-model.h and noticing the include of traced-callback.h and inferring that this must be the file you want. Either way works, slay. 💅
In either case, the next step is to take a look at src/core/model/traced-callback.h in your favorite editor to see what is happening.
You will see a comment at the top of the file that should be comforting:
An ns3::TracedCallback has almost exactly the same API as a normal ns3::Callback but instead of forwarding calls to a single function (as an ns3::Callback normally does), it forwards calls to a chain of ns3::Callback.
This should sound very familiar and let you know you're on the right track. 🎯
Just after this comment, you will find
template<typename T1 = empty, typename T2 = empty,
typename T3 = empty, typename T4 = empty,
typename T5 = empty, typename T6 = empty,
typename T7 = empty, typename T8 = empty>
class TracedCallback
{
...This tells you that TracedCallback is a templated class. It has eight possible type parameters with default values. Go back and compare this with the declaration you are trying to understand:
TracedCallback<Ptr<const MobilityModel> > m_courseChangeTrace;The typename T1 in the templated class declaration corresponds to the Ptr<const MobilityModel> in the declaration above. All of the other type parameters are left as defaults. Looking at the constructor really doesn't tell you much. The one place where you've seen a connection made between your Callback function and the tracing system is in the Connect and ConnectWithoutContext functions. If you scroll down, you'll see a ConnectWithoutContext method here:
template<typename T1, typename T2,
typename T3, typename T4,
typename T5, typename T6,
typename T7, typename T8>
void
TracedCallback<T1,T2,T3,T4,T5,T6,T7,T8>::ConnectWithoutContext ...
{
Callback<void,T1,T2,T3,T4,T5,T6,T7,T8> cb;
cb.Assign (callback);
m_callbackList.push_back (cb);
}You are now in the belly of the beast. 🐉 When the template is instantiated for the declaration above, the compiler will replace T1 with Ptr<const MobilityModel>.
void
TracedCallback<Ptr<const MobilityModel>::ConnectWithoutContext ... cb
{
Callback<void, Ptr<const MobilityModel> > cb;
cb.Assign (callback);
m_callbackList.push_back (cb);
}You can now see the implementation of everything we've been talking about. The code creates a Callback of the right type and assigns your function to it. This is the equivalent of the pfi = MyFunction we discussed at the start of this section. The code then adds the Callback to the list of Callbacks for this source. The only thing left is to look at the definition of Callback. Using the same grep trick as we used to find TracedCallback, you'll be able to find that the file ./core/callback.h is the one we need to look at.
If you look down through the file, you'll see a lot of probably almost incomprehensible template code. You'll eventually come to some API Documentation for the Callback template class, though. Fortunately, there is some English:
Callback template class. This class template implements the Functor Design Pattern. It is used to declare the type of a Callback: - the first non-optional template argument represents the return type of the callback. - the remaining (optional) template arguments represent the type of the subsequent arguments to the callback. - up to nine arguments are supported.
We're trying to figure out what the
Callback<void, Ptr<const MobilityModel> > cb;declaration means. Now we're in a position to understand that the first (non-optional) template argument, void, represents the return type of the Callback. The second (optional) template argument, Ptr<const MobilityModel> represents the type of the first argument to the callback. 🌈
The Callback in question is your function to receive the trace events. From this you can infer that you need a function that returns void and takes a Ptr<const MobilityModel>. For example,
void
CourseChangeCallback (Ptr<const MobilityModel> model)
{
...
}That's all you need if you want to Config::ConnectWithoutContext. If you want a context, you need to Config::Connect and use a Callback function that takes a string context. This is because the Connect function will provide the context for you. You'll need:
void
CourseChangeCallback (std::string context, Ptr<const MobilityModel> model)
{
...
}If you want to ensure that your CourseChangeCallback is only visible in your local file, you can add the keyword static and come up with:
static void
CourseChangeCallback (std::string path, Ptr<const MobilityModel> model)
{
...
}which is exactly what we used in the third.cc example. Maybe you should now go back and reread the previous section (Take My Word for It). 👑
If you're interested in more details regarding the implementation of Callbacks, feel free to take a look at the ns-3 manual. They are one of the most frequently used constructs in the low-level parts of ns-3. It is, in my opinion, a quite elegant thing. ✨
🦋 TracedValues 🦋
Earlier in this section, we presented a simple piece of code that used a TracedValue<int32_t> to demonstrate the basics of the tracing code. We just glossed over what a TracedValue really is and how to find the return type and formal arguments for the callback. 🙈
...and like, bestie, this is where it gets so good, because value semantics is the whole personality of this file. 💅
traced-value.h is the glow-up that pulls in every declaration you need to trace data that obeys value semantics. And what does that even mean, you ask? Lowkey it just means you can pass the object itself around — not the address of the object, the actual object. 🫶 We then stretch that requirement all the way out to cover the whole full set of assignment-style operators that are pre-defined for plain-old-data (POD) types:
operator=(assignment)operator*=,operator/=operator+=,operator-=operator++(both prefix and postfix)operator--(both prefix and postfix)operator<<=,operator>>=operator&=,operator|=operator%=,operator^=
And what that actually means, fr fr, is that you can trace every single change made using those operators to a C++ object that has value semantics. Not one change. All of them. 👑
The TracedValue<> declaration we were just looking at is the infrastructure girlie that overloads all those operators and drives the whole callback process. The second you use any of those operators on a TracedValue, it hands you both the old value and the new value of that variable — in this case an int32_t value. And by just inspecting the TracedValue declaration, we already know the trace sink function is gonna have arguments (int32_t oldValue, int32_t newValue). The return type for a TracedValue callback function is always void, so the expected callback signature for the sink function traceSink is:
void (* traceSink)(int32_t oldValue, int32_t newValue);Now the .AddTraceSource in the GetTypeId method is what provides the "hooks" used for connecting the trace source to the outside world through the Config system. We already talked about the first three arguments to AddTraceSource: the Attribute name for the Config system, a help string, and the address of the TracedValue class data member. ✨
And that final string argument — "ns3::TracedValueCallback::Int32" in the example — is the name of a typedef for the callback function signature. We require these signatures to be defined, and we hand the fully qualified type name to AddTraceSource, so that the API documentation can link a trace source to the function signature. For TracedValue the signature is straightforward; for TracedCallbacks we've already seen that the API docs really help. 🎀
🌈 Real Example 🌈
Okay so let's do an example taken from one of the best-known books on TCP that exists. "TCP/IP Illustrated, Volume 1: The Protocols," by W. Richard Stevens is a literal classic, no cap. I just flipped the book open and ran across a nice plot of both the congestion window and sequence numbers versus time on page 366. Stevens calls this, "Figure 21.10. Value of cwnd and send sequence number while data is being transmitted." So let's just recreate the cwnd part of that plot in ns-3 using the tracing system and gnuplot. 🦋
🌸 Available Sources 🌸
First thing to think about is how we want to get the data out. What is it that we need to trace? So let's consult the "All Trace Sources" list to see what we have to work with. Recall that this is found in the ns-3 API Documentation. If you scroll through the list, you will eventually find:
ns3::TcpSocketBase
- CongestionWindow: The TCP connection's congestion window
- SlowStartThreshold: TCP slow start threshold (bytes)
Turns out the ns-3 TCP implementation lives (mostly) in the file src/internet/model/tcp-socket-base.cc while congestion control variants are in files such as src/internet/model/tcp-bic.cc. And if you don't know this a priori? Bestie, that's what the recursive grep trick is for:
$ find . -name '*.cc' | xargs grep -i tcpYou will find page after page of instances of tcp pointing you to that file. 💀
Bringing up the class documentation for TcpSocketBase and skipping to the list of TraceSources you will find
TraceSources
- CongestionWindow: The TCP connection's congestion window
Callback signature: ns3::TracedValueCallback::Uint32
Clicking on the callback typedef link we see the signature you now know to expect:
typedef void(* ns3::TracedValueCallback::Int32)(int32_t oldValue, int32_t newValue)You should now understand this code completely. If we have a pointer to the TcpSocketBase object, we can TraceConnect to the "CongestionWindow" trace source if we provide an appropriate callback target. This is the same kind of trace source that we saw in the simple example at the start of this section, except that we are talking about uint32_t instead of int32_t. And we know that we have to provide a callback function with that signature. Periodt. 💅
🎀 Finding Examples 🎀
It's always best to try and find working code laying around that you can modify, rather than starting from scratch. So the first order of business now is to find some code that already hooks the "CongestionWindow" trace source and see if we can modify it. As usual, grep is your friend:
$ find . -name '*.cc' | xargs grep CongestionWindowThis will point out a couple of promising candidates: examples/tcp/tcp-large-transfer.cc and src/test/ns3tcp/ns3tcp-cwnd-test-suite.cc.
We haven't visited any of the test code yet, so let's take a look there. You will typically find that test code is fairly minimal, so this is probably a very good bet. Open src/test/ns3tcp/ns3tcp-cwnd-test-suite.cc in your favorite editor and search for "CongestionWindow". You will find,
ns3TcpSocket->TraceConnectWithoutContext ("CongestionWindow",
MakeCallback (&Ns3TcpCwndTestCase1::CwndChange, this));This should look very familiar to you. We mentioned above that if we had a pointer to the TcpSocketBase, we could TraceConnect to the "CongestionWindow" trace source. That's exactly what we have here; so it turns out that this line of code does exactly what we want. Let's go ahead and extract the code we need from this function (Ns3TcpCwndTestCase1::DoRun (void)). If you look at this function, you will find that it looks just like an ns-3 script. It turns out that is exactly what it is. It is a script run by the test framework, so we can just pull it out and wrap it in main instead of in DoRun. Rather than walk through this, step, by step, we have provided the file that results from porting this test back to a native ns-3 script – examples/tutorial/fifth.cc. 🔥
💜 Dynamic Trace Sources 💜
The fifth.cc example demonstrates an extremely important rule that you must understand before using any kind of trace source: you must ensure that the target of a Config::Connect command exists before trying to use it. This is no different than saying an object must be instantiated before trying to call it. Although this may seem obvious when stated this way, it does trip up many people trying to use the system for the first time. 🙈
Let's return to basics for a moment. There are three basic execution phases that exist in any ns-3 script:
- The first phase is sometimes called "Configuration Time" or "Setup Time," and exists during the period when the
mainfunction of your script is running, but beforeSimulator::Runis called. - The second phase is sometimes called "Simulation Time" and exists during the time period when
Simulator::Runis actively executing its events. - After it completes executing the simulation,
Simulator::Runwill return control back to themainfunction. When this happens, the script enters what can be called the "Teardown Phase," which is when the structures and objects created during setup are taken apart and released.
Perhaps the most common mistake made in trying to use the tracing system is assuming that entities constructed dynamically during simulation time are available during configuration time. In particular, an ns-3 Socket is a dynamic object often created by Applications to communicate between Nodes. An ns-3 Application always has a "Start Time" and a "Stop Time" associated with it. In the vast majority of cases, an Application will not attempt to create a dynamic object until its StartApplication method is called at some "Start Time". This is to ensure that the simulation is completely configured before the app tries to do anything (what would happen if it tried to connect to a Node that didn't exist yet during configuration time? 😭). As a result, during the configuration phase you can't connect a trace source to a trace sink if one of them is created dynamically during the simulation.
Bestie, this is the trap. Not me watching you try to Config::Connect to a Socket that literally does not exist yet. 🫠
The two solutions to this conundrum are
- Create a simulator event that is run after the dynamic object is created and hook the trace when that event is executed; or
- Create the dynamic object at configuration time, hook it then, and give the object to the system to use during simulation time.
We took the second approach in the fifth.cc example. This decision required us to create the MyApp Application, the entire purpose of which is to take a Socket as a parameter. 👑
🦋 Walkthrough: fifth.cc 🦋
Now, let's take a look at the example program we constructed by dissecting the congestion window test. Open examples/tutorial/fifth.cc in your favorite editor. You should see some familiar looking code:
/* -*- Mode:C++; c-file-style:"gnu"; indent-tabs-mode:nil; -*- */
/*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License version 2 as
* published by the Free Software Foundation;
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
* Foundation, Include., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
*/
#include <fstream>
#include "ns3/core-module.h"
#include "ns3/network-module.h"
#include "ns3/internet-module.h"
#include "ns3/point-to-point-module.h"
#include "ns3/applications-module.h"
using namespace ns3;
NS_LOG_COMPONENT_DEFINE ("FifthScriptExample");okay so all of that ^^^ we already did, bestie, we are NOT rehashing it, we're moving on 💅 Next up in the source file we've got the network illustration plus a comment that deals with the exact Socket problem we were just talking about.
// ===========================================================================
//
// node 0 node 1
// +----------------+ +----------------+
// | ns-3 TCP | | ns-3 TCP |
// +----------------+ +----------------+
// | 10.1.1.1 | | 10.1.1.2 |
// +----------------+ +----------------+
// | point-to-point | | point-to-point |
// +----------------+ +----------------+
// | |
// +---------------------+
// 5 Mbps, 2 ms
//
//
// We want to look at changes in the ns-3 TCP congestion window. We need
// to crank up a flow and hook the CongestionWindow attribute on the socket
// of the sender. Normally one would use an on-off application to generate a
// flow, but this has a couple of problems. First, the socket of the on-off
// application is not created until Application Start time, so we wouldn't be
// able to hook the socket (now) at configuration time. Second, even if we
// could arrange a call after start time, the socket is not public so we
// couldn't get at it.
//
// So, we can cook up a simple version of the on-off application that does what
// we want. On the plus side we don't need all of the complexity of the on-off
// application. On the minus side, we don't have a helper, so we have to get
// a little more involved in the details, but this is trivial.
//
// So first, we create a socket and do the trace connect on it; then we pass
// this socket into the constructor of our simple application which we then
// install in the source node.
// ===========================================================================
//this one basically explains itself, no notes 💖
Next bit is the declaration of the MyApp Application — the little custom app we threw together so the Socket actually exists at configuration time.
class MyApp : public Application
{
public:
MyApp ();
virtual ~MyApp();
void Setup (Ptr<Socket> socket, Address address, uint32_t packetSize,
uint32_t nPackets, DataRate dataRate);
private:
virtual void StartApplication (void);
virtual void StopApplication (void);
void ScheduleTx (void);
void SendPacket (void);
Ptr<Socket> m_socket;
Address m_peer;
uint32_t m_packetSize;
uint32_t m_nPackets;
DataRate m_dataRate;
EventId m_sendEvent;
bool m_running;
uint32_t m_packetsSent;
};So you can see this class inherits from the ns-3 Application class, fr fr. If you're curious what exactly got inherited, go peek at src/network/model/application.h. And because of that inheritance, MyApp is obligated to override the StartApplication and StopApplication methods. Those two get called automatically whenever MyApp is supposed to start and stop sending data during the simulation. no cap, they just show up. 🦋
✨ Starting/Stopping Applications ✨
okay bestie, this part is a lil deep and I'm not gonna lie to you. It's genuinely worth spending a minute explaining how events actually get started in this system. It's a pretty deep dive, so you can 100% skip it if you're not planning on crawling down into the guts of everything. But it's useful, because the discussion touches on how some very important parts of ns-3 work, and it exposes some important idioms. So if you're planning on implementing new models? Yeah, you probably want to understand this section. 💅
The most common way to start pumping events is to start an Application. That happens as the result of these (hopefully) familiar lines of an ns-3 script:
ApplicationContainer apps = ...
apps.Start (Seconds (1.0));
apps.Stop (Seconds (10.0));The application container code (see src/network/helper/application-container.h if you're interested) loops through its contained applications and calls:
app->SetStartTime (startTime);as a result of the apps.Start call, and:
app->SetStopTime (stopTime);as a result of the apps.Stop call.
The ultimate result of these calls is that we want the simulator to automatically make calls into our Applications, to tell them when to start and when to stop. In the case of MyApp, it inherits from class Application and overrides StartApplication and StopApplication. Those are the functions the simulator will call at the appropriate time. In the case of MyApp you'll find that MyApp::StartApplication does the initial Bind and Connect on the socket, and then starts data flowing by calling MyApp::SendPacket. And MyApp::StopApplication stops generating packets by cancelling any pending send events, then closes the socket. 🎀
One of the nice things about ns-3 is that you can completely ignore the implementation details of how your Application gets "automagically" called by the simulator at the correct time. But since we've already ventured deep into ns-3 at this point... let's go for it. 🕳️🐇
If you look at src/network/model/application.cc you'll find that the SetStartTime method of an Application just sets the member variable m_startTime, and the SetStopTime method just sets m_stopTime. And from there, without some hints, the trail will probably go cold on you. 😭
The key to picking the trail back up is knowing that there's a global list of all the nodes in the system. Whenever you create a node in a simulation, a pointer to that Node gets added to the global NodeList.
Take a look at src/network/model/node-list.cc and search for NodeList::Add. The public static implementation calls into a private implementation called NodeListPriv::Add. This is a relatively common idom in ns-3. So go look at NodeListPriv::Add. There you'll find:
Simulator::ScheduleWithContext (index, TimeStep (0), &Node::Initialize, node);Which tells you that whenever a Node is created in a simulation, as a side-effect, a call to that node's Initialize method gets scheduled for you, and it happens at time zero. Don't read too much into that name yet. It does not mean the Node is going to start doing anything — it can be interpreted as an informational call into the Node telling it that the simulation has started, not a call for action telling the Node to start doing something. 💅
So NodeList::Add indirectly schedules a call to Node::Initialize at time zero, to advise a new Node that the simulation has started. If you look in src/network/model/node.h, you will, however, not find a method called Node::Initialize. Turns out the Initialize method is inherited from class Object. All objects in the system can be notified when the simulation starts, and objects of class Node are just one kind of those objects. 🫶
Next, take a look at src/core/model/object.cc and search for Object::Initialize. This code is not as straightforward as you might have expected, since ns-3 Objects support aggregation. The code in Object::Initialize then loops through all of the objects that have been aggregated together and calls their DoInitialize method. This is another idiom that's very common in ns-3, sometimes called the "template design pattern": a public non-virtual API method, which stays constant across implementations, and that calls a private virtual implementation method that is inherited and implemented by subclasses. The names are typically something like MethodName for the public API and DoMethodName for the private API. 🎀
This tells us we should go looking for a Node::DoInitialize method in src/network/model/node.cc — that's the method that'll continue our trail. If you locate the code, you'll find a method that loops through all of the devices in the Node and then all of the applications in the Node, calling device->Initialize and application->Initialize respectively.
You may already know that classes Device and Application both inherit from class Object, so the next step is to look at what happens when Application::DoInitialize gets called. Take a look at src/network/model/application.cc and you'll find:
void
Application::DoInitialize (void)
{
m_startEvent = Simulator::Schedule (m_startTime, &Application::StartApplication, this);
if (m_stopTime != TimeStep (0))
{
m_stopEvent = Simulator::Schedule (m_stopTime, &Application::StopApplication, this);
}
Object::DoInitialize ();
}see, told you we'd get there 💖
okay bestie, we have finally reached the end of the trail 🏁 and if you kept all of it straight in your head, that's genuinely icon behavior, no cap. So here's the tea: when you implement an ns-3 Application, your shiny new application inherits from class Application. You override the StartApplication and StopApplication methods and provide mechanisms for starting and stopping the flow of data out of your new Application. ✨
When a Node is created in the simulation, it gets added to a global NodeList. The act of adding a Node to this NodeList causes a simulator event to be scheduled for time zero which calls the Node::Initialize method of the newly added Node to be called when the simulation starts. Since a Node inherits from Object, this calls the Object::Initialize method on the Node which, in turn, calls the DoInitialize methods on all of the Objects aggregated to the Node (think mobility models, bestie 🛼). Since the Node Object has overridden DoInitialize, that method is called when the simulation starts. The Node::DoInitialize method calls the Initialize methods of all of the Applications on the node. Since Applications are also Objects, this causes Application::DoInitialize to be called. When Application::DoInitialize is called, it schedules events for the StartApplication and StopApplication calls on the Application. These calls are designed to start and stop the flow of data from the Application. 🫶
This has been another fairly long journey, but it only has to be made once, and you now understand another very deep piece of ns-3. periodt. 💅
✨ The MyApp Application ✨
The MyApp Application needs a constructor and a destructor, of course 🎀:
MyApp::MyApp ()
: m_socket (0),
m_peer (),
m_packetSize (0),
m_nPackets (0),
m_dataRate (0),
m_sendEvent (),
m_running (false),
m_packetsSent (0)
{
}
MyApp::~MyApp()
{
m_socket = 0;
}The existence of the next bit of code is the whole reason why we wrote this Application in the first place, bestie. it's giving purpose. 🦋
void
MyApp::Setup (Ptr<Socket> socket, Address address, uint32_t packetSize,
uint32_t nPackets, DataRate dataRate)
{
m_socket = socket;
m_peer = address;
m_packetSize = packetSize;
m_nPackets = nPackets;
m_dataRate = dataRate;
}This code should be pretty self-explanatory. We are just initializing member variables, nothing dramatic. The important one from the perspective of tracing is the Ptr<Socket> socket which we needed to provide to the application during configuration time. Recall that we are going to create the Socket as a TcpSocket (which is implemented by TcpSocketBase) and hook its “CongestionWindow” trace source before passing it to the Setup method. 🎯
void
MyApp::StartApplication (void)
{
m_running = true;
m_packetsSent = 0;
m_socket->Bind ();
m_socket->Connect (m_peer);
SendPacket ();
}The above code is the overridden implementation Application::StartApplication that will be automatically called by the simulator to start our Application running at the appropriate time. You can see that it does a Socket Bind operation. If you are familiar with Berkeley Sockets this shouldn't be a surprise — it performs the required work on the local side of the connection just as you might expect. 👑
The following Connect will do what is required to establish a connection with the TCP at Address m_peer. It should now be clear why we need to defer a lot of this to simulation time, since the Connect is going to need a fully functioning network to complete. After the Connect, the Application then starts creating simulation events by calling SendPacket. 🌸
The next bit of code explains to the Application how to stop creating simulation events:
void
MyApp::StopApplication (void)
{
m_running = false;
if (m_sendEvent.IsRunning ())
{
Simulator::Cancel (m_sendEvent);
}
if (m_socket)
{
m_socket->Close ();
}
}Every time a simulation event is scheduled, an Event is created. If the Event is pending execution or executing, its method IsRunning will return true. In this code, if IsRunning() returns true, we Cancel the event which removes it from the simulator event queue. By doing this, we break the chain of events that the Application is using to keep sending its Packets and the Application goes quiet 🤫. After we quiet the Application we Close the socket which tears down the TCP connection. bye felicia. 💀
The socket is actually deleted in the destructor when the m_socket = 0 is executed. This removes the last reference to the underlying Ptr<Socket> which causes the destructor of that Object to be called. lowkey poetic. ✨
Recall that StartApplication called SendPacket to start the chain of events that describes the Application behavior:
void
MyApp::SendPacket (void)
{
Ptr<Packet> packet = Create<Packet> (m_packetSize);
m_socket->Send (packet);
if (++m_packetsSent < m_nPackets)
{
ScheduleTx ();
}
}Here, you see that SendPacket does just that. It creates a Packet and then does a Send which, if you know Berkeley Sockets, is probably just what you expected to see. ate. 🔥
It is the responsibility of the Application to keep scheduling the chain of events, so the next lines call ScheduleTx to schedule another transmit event (a SendPacket) until the Application decides it has sent enough. 🐝
void
MyApp::ScheduleTx (void)
{
if (m_running)
{
Time tNext (Seconds (m_packetSize * 8 / static_cast<double> (m_dataRate.GetBitRate ())));
m_sendEvent = Simulator::Schedule (tNext, &MyApp::SendPacket, this);
}
}Here, you see that ScheduleTx does exactly that. If the Application is running (if StopApplication has not been called) it will schedule a new event, which calls SendPacket again.
🚨 bestie, the alert reader will spot something that also trips up new users. The data rate of an Application is just that. It has nothing to do with the data rate of an underlying Channel. This is the rate at which the Application produces bits. It does not take into account any overhead for the various protocols or channels that it uses to transport the data. If you set the data rate of an Application to the same data rate as your underlying Channel you will eventually get a buffer overflow. not cute. 😭
✨ Trace Sinks ✨
The whole point of this exercise is to get trace callbacks from TCP indicating the congestion window has been updated. The next piece of code implements the corresponding trace sink 💧:
static void
CwndChange (uint32_t oldCwnd, uint32_t newCwnd)
{
NS_LOG_UNCOND (Simulator::Now ().GetSeconds () << "\t" << newCwnd);
}This should be very familiar to you now, so we won't dwell on the details. This function just logs the current simulation time and the new value of the congestion window every time it is changed. You can probably imagine that you could load the resulting output into a graphics program (gnuplot or Excel) and immediately see a nice graph of the congestion window behavior over time 📈. main character data visualization energy.
We added a new trace sink to show where packets are dropped. We are going to add an error model to this code also, so we wanted to demonstrate this working. 🫧
static void
RxDrop (Ptr<const Packet> p)
{
NS_LOG_UNCOND ("RxDrop at " << Simulator::Now ().GetSeconds ());
}This trace sink will be connected to the “PhyRxDrop” trace source of the point-to-point NetDevice. This trace source fires when a packet is dropped by the physical layer of a NetDevice. If you take a small detour to the source (src/point-to-point/model/point-to-point-net-device.cc) you will see that this trace source refers to PointToPointNetDevice::m_phyRxDropTrace. If you then look in src/point-to-point/model/point-to-point-net-device.h for this member variable, you will find that it is declared as a TracedCallback<Ptr<const Packet> >. This should tell you that the callback target should be a function that returns void and takes a single parameter which is a Ptr<const Packet> (assuming we use ConnectWithoutContext) – just what we have above. iykyk. 💜
✨ Main Program ✨
The following code should be very familiar to you by now 🌈:
int
main (int argc, char *argv[])
{
NodeContainer nodes;
nodes.Create (2);
PointToPointHelper pointToPoint;
pointToPoint.SetDeviceAttribute ("DataRate", StringValue ("5Mbps"));
pointToPoint.SetChannelAttribute ("Delay", StringValue ("2ms"));
NetDeviceContainer devices;
devices = pointToPoint.Install (nodes);This creates two nodes with a point-to-point channel between them, just as shown in the illustration at the start of the file. two little nodes, living their best lives, connected. 🎀
The next few lines of code show something new. If we trace a connection that behaves perfectly, we will end up with a monotonically increasing congestion window. To see any interesting behavior, we really want to introduce link errors which will drop packets, cause duplicate ACKs and trigger the more interesting behaviors of the congestion window. chaos is the point, bestie. 💅
ns-3 provides ErrorModel objects which can be attached to Channels. We are using the RateErrorModel which allows us to introduce errors into a Channel at a given rate. 🌧️
Ptr<RateErrorModel> em = CreateObject<RateErrorModel> ();
em->SetAttribute ("ErrorRate", DoubleValue (0.00001));
devices.Get (1)->SetAttribute ("ReceiveErrorModel", PointerValue (em));The above code instantiates a RateErrorModel Object, and we set the “ErrorRate” Attribute to the desired value. We then set the resulting instantiated RateErrorModel as the error model used by the point-to-point NetDevice. This will give us some retransmissions and make our plot a little more interesting. ✨ spicy packet loss ✨
InternetStackHelper stack;
stack.Install (nodes);
Ipv4AddressHelper address;
address.SetBase ("10.1.1.0", "255.255.255.252");
Ipv4InterfaceContainer interfaces = address.Assign (devices);The above code should be familiar. It installs internet stacks on our two nodes and creates interfaces and assigns IP addresses for the point-to-point devices. 🦋
Since we are using TCP, we need something on the destination Node to receive TCP connections and data. The PacketSink Application is commonly used in ns-3 for that purpose. 👑
uint16_t sinkPort = 8080;
Address sinkAddress (InetSocketAddress(interfaces.GetAddress (1), sinkPort));
PacketSinkHelper packetSinkHelper ("ns3::TcpSocketFactory",
InetSocketAddress (Ipv4Address::GetAny (), sinkPort));
ApplicationContainer sinkApps = packetSinkHelper.Install (nodes.Get (1));
sinkApps.Start (Seconds (0.));
sinkApps.Stop (Seconds (20.));Okay bestie, this whole chunk is giving ✨ déjà vu ✨ — you've literally seen all of it before. All of it except one lil line:
PacketSinkHelper packetSinkHelper ("ns3::TcpSocketFactory",
InetSocketAddress (Ipv4Address::GetAny (), sinkPort));What's it doing? It's instantiating a PacketSinkHelper and telling it to make sockets with the class ns3::TcpSocketFactory. That class implements a design pattern called an "object factory" — a super commonly used mechanism for saying which class gets used to create objects, but in an abstract way, y'know? So here, instead of you having to make the objects yourself, you hand the PacketSinkHelper a string that specifies a TypeId string. That string is used to create an object, and that object then gets used, in turn, to create instances of the Objects that the factory makes. Layers, babe. 💅
The other parameter tells the Application which address and port it should Bind to.
Next two lines make the socket and hook up the trace source:
Ptr<Socket> ns3TcpSocket = Socket::CreateSocket (nodes.Get (0),
TcpSocketFactory::GetTypeId ());
ns3TcpSocket->TraceConnectWithoutContext ("CongestionWindow",
MakeCallback (&CwndChange));The first statement calls the static member function Socket::CreateSocket and hands it a Node plus an explicit TypeId for the object factory that's gonna create the socket. This is a slightly lower level call than the PacketSinkHelper call up above — it uses an explicit C++ type instead of one referred to by a string. Otherwise? Conceptually the exact same thing, no cap.
Once the TcpSocket is created and attached to the Node, we can use TraceConnectWithoutContext to connect the CongestionWindow trace source to our trace sink. 🎀
Real talk, remember how we coded an Application so we could take that Socket we just made (during configuration time) and use it in simulation time? Yeah. So now we gotta instantiate that Application. We didn't go to any trouble to create a helper to manage the Application, so we're gonna have to create and install it "manually". And honestly? It's actually quite easy:
Ptr<MyApp> app = CreateObject<MyApp> ();
app->Setup (ns3TcpSocket, sinkAddress, 1040, 1000, DataRate ("1Mbps"));
nodes.Get (0)->AddApplication (app);
app->Start (Seconds (1.));
app->Stop (Seconds (20.));Line one creates an Object of type MyApp — that's our Application. Line two tells the Application what Socket to use, what address to connect to, how much data to send at each send event, how many send events to generate, and the rate at which to produce data from those events. That's a lot of responsibility for one line, she's booked and busy fr fr. 🫶
Next, we manually add the MyApp Application to the source Node and explicitly call the Start and Stop methods on the Application to tell it when to start and stop doing its thing.
Now we need to actually do the connect from the receiver point-to-point NetDevice drop event to our RxDrop callback:
devices.Get (1)->TraceConnectWithoutContext("PhyRxDrop", MakeCallback (&RxDrop));At this point it should be obvious bestie — we're grabbing a reference to the receiving Node NetDevice from its container and connecting the trace source defined by the attribute "PhyRxDrop" on that device to the trace sink RxDrop. Easy. Iconic. 💖
Finally, we tell the simulator to override any Applications and just stop processing events at 20 seconds into the simulation.
Simulator::Stop (Seconds(20));
Simulator::Run ();
Simulator::Destroy ();
return 0;
}Remember: the second Simulator::Run gets called, configuration time ends and simulation time begins. All of that work we orchestrated by creating the Application and teaching it how to connect and send data — all of it actually happens during this one function call. 😭
And the second Simulator::Run returns, the simulation is complete and we enter the teardown phase. In this case, Simulator::Destroy takes care of the gory details, and we just return a success code after it completes.
🌸 Running fifth.cc 🌸
Since we've provided the file fifth.cc for you, if you've built your distribution (in debug mode since it uses NS_LOG — recall that optimized builds optimize out NS_LOG) it'll be sitting there waiting for you to run it. 💅
$ ./waf --run fifth
Waf: Entering directory `/home/craigdo/repos/ns-3-allinone-dev/ns-3-dev/build'
Waf: Leaving directory `/home/craigdo/repos/ns-3-allinone-dev/ns-3-dev/build'
'build' finished successfully (0.684s)
1 536
1.0093 1072
1.01528 1608
1.02167 2144
...
1.11319 8040
1.12151 8576
1.12983 9112
RxDrop at 1.13696
...Okay you can probably immediately clock a downside of using prints of any kind in your traces. 💀 We get those extraneous waf messages printed all over our interesting information, plus those RxDrop messages. We'll remedy that soon, but I know you cannot wait to see the results of all this work. So let's redirect that output to a file called cwnd.dat:
$ ./waf --run fifth > cwnd.dat 2>&1Now edit up cwnd.dat in your favorite editor and remove the waf build status and drop lines, leaving only the traced data. (You could also just as easily comment out the TraceConnectWithoutContext("PhyRxDrop", MakeCallback (&RxDrop)); in the script to get rid of the drop prints. Same energy. 😌)
You can now run gnuplot (if you have it installed) and tell it to generate some pretty pictures:
$ gnuplot
gnuplot> set terminal png size 640,480
gnuplot> set output "cwnd.png"
gnuplot> plot "cwnd.dat" using 1:2 title 'Congestion Window' with linespoints
gnuplot> exitYou should now have a graph of the congestion window versus time sitting in the file cwnd.png that looks like:
[image]🦋 Using Mid-Level Helpers 🦋
In the previous section, we showed how to hook a trace source and get (hopefully) interesting information out of a simulation. Now, perhaps you'll recall that we called logging to the standard output using std::cout a "blunt instrument" much earlier in this chapter. We also wrote about how it was a problem having to parse the log output in order to isolate interesting information. And it may have occurred to you that we just spent a lot of time implementing an example that exhibits all of the problems we claim to fix with the ns-3 tracing system! 🙈
You would be correct. But bear with us — we're not done yet. 👑
One of the most important things we want to do is have the ability to easily control the amount of output coming out of the simulation. And we also want to save that data to a file so we can refer back to it later. We can use the mid-level trace helpers provided in ns-3 to do exactly that and complete the picture. ✨
We provide a script that writes the cwnd change and drop events developed in the example fifth.cc to disk in separate files. The cwnd changes are stored as a tab-separated ASCII file, and the drop events are stored in a PCAP file. The changes to make this happen are quite small.
💜 Walkthrough: sixth.cc 💜
Let's take a look at the changes required to go from fifth.cc to sixth.cc. Open examples/tutorial/sixth.cc in your favorite editor. You can see the first change by searching for CwndChange. You'll find that we've changed the signatures for the trace sinks and added a single line to each sink that writes the traced information to a stream representing a file. 🎀
static void
CwndChange (Ptr<OutputStreamWrapper> stream, uint32_t oldCwnd, uint32_t newCwnd)
{
NS_LOG_UNCOND (Simulator::Now ().GetSeconds () << "\t" << newCwnd);
*stream->GetStream () << Simulator::Now ().GetSeconds () << "\t" << oldCwnd << "\t" << newCwnd << std::endl;
}
static void
RxDrop (Ptr<PcapFileWrapper> file, Ptr<const Packet> p)
{
NS_LOG_UNCOND ("RxDrop at " << Simulator::Now ().GetSeconds ());
file->Write(Simulator::Now(), p);
}We've added a "stream" parameter to the CwndChange trace sink. This is an object that holds — and keeps safely alive — a C++ output stream. Turns out it's a very simple object, but one that manages lifetime issues for the stream and solves a problem that even experienced C++ users run into. 😭
Here's the tea: the copy constructor for std::ostream is marked private. Which means std::ostreams do not obey value semantics and cannot be used in any mechanism that requires the stream to be copied. And that includes the ns-3 callback system, which — as you may recall — requires objects that obey value semantics. So yeah. That's why this lil wrapper exists. 💅
Also notice we added the following line in the CwndChange trace sink implementation:
*stream->GetStream () << Simulator::Now ().GetSeconds () << "\t" << oldCwnd << "\t" << newCwnd << std::endl;This would be very familiar code if you replaced *stream->GetStream () with std::cout, like:
std::cout << Simulator::Now ().GetSeconds () << "\t" << oldCwnd << "\t" << newCwnd << std::endl;Which illustrates that the Ptr<OutputStreamWrapper> is really just carrying around a std::ofstream for you, and you can use it here like any other output stream. Slay. 🫶
A similar situation happens in RxDrop, except the object being passed around (a Ptr<PcapFileWrapper>) represents a PCAP file. There's a one-liner in the trace sink to write a timestamp and the contents of the packet being dropped to the PCAP file:
file->Write(Simulator::Now(), p);And of course, if we have objects representing the two files, we need to create them somewhere and also cause them to be passed to the trace sinks. If you look in the main function, you'll find new code doing exactly that:
AsciiTraceHelper asciiTraceHelper;
Ptr<OutputStreamWrapper> stream = asciiTraceHelper.CreateFileStream ("sixth.cwnd");
ns3TcpSocket->TraceConnectWithoutContext ("CongestionWindow", MakeBoundCallback (&CwndChange, stream));
...
PcapHelper pcapHelper;
Ptr<PcapFileWrapper> file = pcapHelper.CreateFile ("sixth.pcap", std::ios::out, PcapHelper::DLT_PPP);
devices.Get (1)->TraceConnectWithoutContext("PhyRxDrop", MakeBoundCallback (&RxDrop, file));In the first section of the code snippet above, we're creating the ASCII trace file, creating an object responsible for managing it, and using a variant of the callback creation function to arrange for the object to be passed to the sink. Our ASCII trace helpers provide a rich set of functions to make using text (ASCII) files easy — but here we're just gonna illustrate the use of the file stream creation function. 🦋
The CreateFileStream function is basically gonna instantiate a std::ofstream object and create a brand new file (or truncate one that already exists, rip to whatever was in there 💀). That std::ofstream gets packaged up inside an ns-3 object so somebody's managing its lifetime and so we don't get tangled up in copy constructor drama. Iconic behavior honestly.
Then we take that ns-3 object that represents the file and hand it over to MakeBoundCallback(). This function makes a callback just like MakeCallback() does, but it “binds” a new value to the callback. That value gets tacked on as the first argument to the callback before it ever gets called.
Basically, MakeBoundCallback(&CwndChange, stream) makes the trace source shove that extra stream parameter onto the front of the formal parameter list before it invokes the callback. Which means the signature of the CwndChange sink has to change to match the one we showed up above — the one that includes the "extra" parameter Ptr<OutputStreamWrapper> stream. No cap, that param isn't optional.
In the second chunk of code in the snippet above, we instantiate a PcapHelper to do the exact same trick for our PCAP trace file that we did with the AsciiTraceHelper. This line right here:
Ptr<PcapFileWrapper> file = pcapHelper.CreateFile ("sixth.pcap", "w", PcapHelper::DLT_PPP);creates a PCAP file named sixth.pcap with file mode "w". And mode "w" means the new file gets truncated — contents deleted, gone, bye — if a file with that name already exists. The final parameter is the “data link type” of the new PCAP file. These are the same as the PCAP library data link types defined in bpf.h, if you happen to be familiar with PCAP like that. In this case, DLT_PPP says the PCAP file is gonna hold packets prefixed with point to point headers. Which is true, because the packets are coming from our point-to-point device driver. Other common data link types:
DLT_EN10MB— that's 10 MB Ethernet, appropriate for csma devices 🌐DLT_IEEE802_11— IEEE 802.11, appropriate for wifi devices 📶
These live in src/network/helper/trace-helper.h if you wanna go peek at the list yourself. The entries in that list match the ones in bpf.h, but ns-3 duplicates them on purpose to avoid a PCAP source dependence. Smart, honestly.
A ns-3 object representing the PCAP file comes back from CreateFile and gets used in a bound callback exactly the same way as in the ASCII case. Same energy, different file format. 💅
🫶 An important detour (please read this one bestie)
It matters a lot to notice that even though both of these objects are declared in very similar-looking ways —
Ptr<PcapFileWrapper> file ...
Ptr<OutputStreamWrapper> stream ...— the underlying objects are entirely different. Like, not even the same species.
- The
Ptr<PcapFileWrapper>is a smart pointer to an ns-3 Object, which is a fairly heavyweight thing — it supports Attributes and it's integrated into the Config system. - The
Ptr<OutputStreamWrapper>, on the other hand, is a smart pointer to a reference counted object that is a very lightweight thing.
So remember: look at the object you're referencing before you make any assumptions about the “powers” that object may have. 🙈
For example, go take a look at src/network/utils/pcap-file-wrapper.h in the distribution and notice:
class PcapFileWrapper : public ObjectSee that? Class PcapFileWrapper is an ns-3 Object purely by virtue of its inheritance. Then look at src/network/model/output-stream-wrapper.h and notice:
class OutputStreamWrapper : public SimpleRefCount<OutputStreamWrapper>That object is not an ns-3 Object at all — it's “merely” a C++ object that just happens to support intrusive reference counting. That's it, that's the whole thing.
The point here, and I need you to internalize this: just because you read Ptr<something> does not necessarily mean that something is an ns-3 Object that you can hang ns-3 Attributes on. Fr fr. ⭐
Back to the example
Now, back to the example. If you build and run this one:
$ ./waf --run sixthyou'll see the same messages show up as when you ran fifth — but two new files will appear in the top-level directory of your ns-3 distribution:
sixth.cwnd sixth.pcapSince sixth.cwnd is an ASCII text file, you can view it with cat or whatever your favorite file viewer is. Look at this gorgeousness:
1 0 536
1.0093 536 1072
1.01528 1072 1608
1.02167 1608 2144
...
9.69256 5149 5204
9.89311 5204 5259You've got a tab separated file with a timestamp, an old congestion window and a new congestion window — perfect for importing directly into your plot program. There are no extraneous prints in the file. No parsing required. No editing required. It just works. 🎀
And since sixth.pcap is a PCAP file, you can view it with tcpdump:
reading from file sixth.pcap, link-type PPP (PPP)
1.136956 IP 10.1.1.1.49153 > 10.1.1.2.8080: Flags [.], seq 17177:17681, ack 1, win 32768, options [TS val 1133 ecr 1127,eol], length 504
1.403196 IP 10.1.1.1.49153 > 10.1.1.2.8080: Flags [.], seq 33280:33784, ack 1, win 32768, options [TS val 1399 ecr 1394,eol], length 504
...
7.426220 IP 10.1.1.1.49153 > 10.1.1.2.8080: Flags [.], seq 785704:786240, ack 1, win 32768, options [TS val 7423 ecr 7421,eol], length 536
9.630693 IP 10.1.1.1.49153 > 10.1.1.2.8080: Flags [.], seq 882688:883224, ack 1, win 32768, options [TS val 9620 ecr 9618,eol], length 536You've got a PCAP file containing the packets that were dropped in the simulation. There are no other packets present in the file, and there's nothing else hanging around in there to make your life difficult. So clean. So tidy. ✨
It's been a long journey, bestie, but we are now at a point where we can genuinely appreciate the ns-3 tracing system. We pulled important events out of the middle of a TCP implementation and out of a device driver. We stored those events directly in files that are usable with commonly known tools. And we did all of it without modifying any of the core code involved — in only 18 lines of code:
static void
CwndChange (Ptr<OutputStreamWrapper> stream, uint32_t oldCwnd, uint32_t newCwnd)
{
NS_LOG_UNCOND (Simulator::Now ().GetSeconds () << "\t" << newCwnd);
*stream->GetStream () << Simulator::Now ().GetSeconds () << "\t" << oldCwnd << "\t" << newCwnd << std::endl;
}
...
AsciiTraceHelper asciiTraceHelper;
Ptr<OutputStreamWrapper> stream = asciiTraceHelper.CreateFileStream ("sixth.cwnd");
ns3TcpSocket->TraceConnectWithoutContext ("CongestionWindow", MakeBoundCallback (&CwndChange, stream));
...
static void
RxDrop (Ptr<PcapFileWrapper> file, Ptr<const Packet> p)
{
NS_LOG_UNCOND ("RxDrop at " << Simulator::Now ().GetSeconds ());
file->Write(Simulator::Now(), p);
}
...
PcapHelper pcapHelper;
Ptr<PcapFileWrapper> file = pcapHelper.CreateFile ("sixth.pcap", "w", PcapHelper::DLT_PPP);
devices.Get (1)->TraceConnectWithoutContext("PhyRxDrop", MakeBoundCallback (&RxDrop, file));18 lines. That's it. Main character energy honestly. 👑
🌈 Trace Helpers 🌈
The ns-3 trace helpers give you a whole rich environment for configuring and selecting different trace events and writing them out to files. In previous sections — primarily BuildingTopologies — we've already seen several varieties of the trace helper methods designed for use inside other (device) helpers.
Maybe you'll recall seeing some of these variations:
pointToPoint.EnablePcapAll ("second");
pointToPoint.EnablePcap ("second", p2pNodes.Get (0)->GetId (), 0);
csma.EnablePcap ("third", csmaDevices.Get (0), true);
pointToPoint.EnableAsciiAll (ascii.CreateFileStream ("myfirst.tr"));What might not be obvious yet, though, is that there's a consistent model behind all of the trace-related methods found in the system. So let's take a little time and look at the “big picture”. 🖼️
There are currently two primary use cases of the tracing helpers in ns-3: device helpers and protocol helpers.
- Device helpers tackle the problem of specifying which traces should be enabled through a (node, device) pair. Like, maybe you want to say that PCAP tracing should be enabled on a particular device on a specific node. This follows from the ns-3 device conceptual model, and also from the conceptual models of the various device helpers. And naturally following from that, the files created follow a
<prefix>-<node>-<device>naming convention. - Protocol helpers tackle the problem of specifying which traces should be enabled through a protocol and interface pair. This follows from the ns-3 protocol stack conceptual model, and also from the conceptual models of internet stack helpers. And naturally, the trace files should follow a
<prefix>-<protocol>-<interface>naming convention.
So the trace helpers fall naturally into a two-dimensional taxonomy. There are subtleties that stop all four classes from behaving identically, but we do strive to make them all work as similarly as possible — and whenever it's possible, there are analogs for all methods in all classes. 🦋
Here's the whole grid:
- Device Helper — PCAP ✓, ASCII ✓
- Protocol Helper — PCAP ✓, ASCII ✓
(That's the \checkmark situation. All four boxes are checked, bestie.)
We use an approach called a mixin to add tracing functionality to our helper classes. A mixin is a class that provides functionality when it is inherited by a subclass. Inheriting from a mixin is not considered a form of specialization — it's really just a way to collect functionality.
Let's take a quick look at all four of these cases and their respective mixins. 💖
💅 Device Helpers
🎀 PCAP
The goal of these helpers is to make it easy to add a consistent PCAP trace facility to an ns-3 device. We want all of the various flavors of PCAP tracing to work the same across all devices, so the methods of these helpers get inherited by device helpers. Go take a look at src/network/helper/trace-helper.h if you want to follow along with the discussion while looking at real code.
The class PcapHelperForDevice is a mixin that provides the high level functionality for using PCAP tracing in an ns-3 device. Every device must implement a single virtual method inherited from this class:
virtual void EnablePcapInternal (std::string prefix, Ptr<NetDevice> nd, bool promiscuous, bool explicitFilename) = 0;The signature of this method reflects the device-centric view of the situation at this level. All of the public methods inherited from class PcapUserHelperForDevice reduce to calling this single device-dependent implementation method. For example, the lowest level PCAP method,
will call the device implementation of EnablePcapInternal directly. Every other public PCAP tracing method is built on top of this one implementation to hand you extra user-level functionality. What that means for you, bestie: every device helper in the system gets all the PCAP trace methods, and they'll all behave the same way across devices — as long as the device implements EnablePcapInternal correctly. 💅
✨ Methods ✨
void EnablePcap (std::string prefix, Ptr<NetDevice> nd, bool promiscuous = false, bool explicitFilename = false);
void EnablePcap (std::string prefix, std::string ndName, bool promiscuous = false, bool explicitFilename = false);
void EnablePcap (std::string prefix, NetDeviceContainer d, bool promiscuous = false);
void EnablePcap (std::string prefix, NodeContainer n, bool promiscuous = false);
void EnablePcap (std::string prefix, uint32_t nodeid, uint32_t deviceid, bool promiscuous = false);
void EnablePcapAll (std::string prefix, bool promiscuous = false);Okay so every single one of those methods up there has a default parameter called promiscuous that defaults to false. What's it saying? It's saying don't gather the trace in promiscuous mode. But if you DO want your traces to include all the traffic the device sees (and the device supports promiscuous mode, that's the catch), just add a true to any of those calls. Like:
Ptr<NetDevice> nd;
...
helper.EnablePcap ("prefix", nd, true);boom, promiscuous mode captures enabled on the NetDevice nd. 🎀
The first two methods also carry a default parameter called explicitFilename — we'll get into that below, promise. 🫶
You're encouraged to go peruse the API Documentation for class PcapHelperForDevice to find the full deets on these methods, but to summarize …
- You can enable PCAP tracing on a particular node/net-device pair by handing a
Ptr<NetDevice>to anEnablePcapmethod. ThePtr<Node>is implicit, because the net device has to belong to exactly one Node. Example:
Ptr<NetDevice> nd;
...
helper.EnablePcap ("prefix", nd);- You can also enable PCAP tracing on a particular node/net-device pair by handing a
std::stringthat represents an object name service string to anEnablePcapmethod. ThePtr<NetDevice>gets looked up from that name string. Again, the<Node>is implicit, since the named net device must belong to exactly one Node. Example:
Names::Add ("server" ...);
Names::Add ("server/eth0" ...);
...
helper.EnablePcap ("prefix", "server/ath0");- You can enable PCAP tracing on a whole collection of node/net-device pairs by handing over a
NetDeviceContainer. For eachNetDevicein the container, the type gets checked. For each device of the proper type (same type as what the device helper manages), tracing is enabled. Again the<Node>is implicit — the found net device has to belong to exactly one Node. Example:
NetDeviceContainer d = ...;
...
helper.EnablePcap ("prefix", d);- You can enable PCAP tracing on a collection of node/net-device pairs by handing over a
NodeContainer. For eachNodein theNodeContainer, its attachedNetDevicesget iterated. For eachNetDeviceattached to eachNodein the container, the type of that device gets checked. For each device of the proper type (same type as the device helper manages), tracing is enabled. 🌸
NodeContainer n;
...
helper.EnablePcap ("prefix", n);- You can enable PCAP tracing based on Node ID and device ID too, not just an explicit
Ptr. EveryNodein the system has an integer Node ID, and every device connected to a Node has an integer device ID. ✨
helper.EnablePcap ("prefix", 21, 1);- And finally, you can enable PCAP tracing for all devices in the system that share the same type as what the device helper manages. 👑
helper.EnablePcapAll ("prefix");✨ Filenames ✨
Implicit in all those method descriptions is that the implementation method builds a complete filename. By convention, PCAP traces in the ns-3 system look like:
<prefix>-<node id>-<device id>.pcapLike we said earlier, every Node in the system gets a system-assigned Node id, and every device gets an interface index (also called a device id) relative to its node. So by default, a PCAP trace file created from enabling tracing on the first device of Node 21 with the prefix prefix would be prefix-21-1.pcap. 💾
You can always use the ns-3 object name service to make this clearer. For example, if you assign the name server to Node 21 via the object name service, the resulting PCAP trace file name automatically becomes prefix-server-1.pcap. And if you also assign the name eth0 to the device, your PCAP file name picks that up too and gets called prefix-server-eth0.pcap. iconic, honestly.
And finally — those two methods we mentioned earlier:
void EnablePcap (std::string prefix, Ptr<NetDevice> nd, bool promiscuous = false, bool explicitFilename = false);
void EnablePcap (std::string prefix, std::string ndName, bool promiscuous = false, bool explicitFilename = false);they have that default parameter called explicitFilename. When you set it to true, it disables the automatic filename completion mechanism and lets you create an explicit filename instead. Heads up: this option is only available in the methods that enable PCAP tracing on a single device. 🙈
So for example, if you want a device helper to create a single promiscuous PCAP capture file with a specific name my-pcap-file.pcap on a given device, you could do:
Ptr<NetDevice> nd;
...
helper.EnablePcap ("my-pcap-file.pcap", nd, true, true);The first true enables promiscuous mode traces, and the second one tells the helper to interpret the prefix parameter as a complete filename. 🌈
✨ ASCII ✨
The behavior of the ASCII trace helper mixin is basically the same vibe as the PCAP version. Take a peek at src/network/helper/trace-helper.h if you wanna follow along while looking at real code. 📖
The class AsciiTraceHelperForDevice adds the high-level functionality for using ASCII tracing to a device helper class. Just like the PCAP case, every device must implement a single virtual method inherited from the ASCII trace mixin.
virtual void EnableAsciiInternal (Ptr<OutputStreamWrapper> stream,
std::string prefix,
Ptr<NetDevice> nd,
bool explicitFilename) = 0;The signature of this method reflects the device-centric view of things at this level — and also the fact that the helper may be writing to a shared output stream. All the public ASCII-trace-related methods inherited from class AsciiTraceHelperForDevice reduce down to calling this one single device-dependent implementation method. For example, the lowest level ascii trace methods:
void EnableAscii (std::string prefix, Ptr<NetDevice> nd, bool explicitFilename = false);
void EnableAscii (Ptr<OutputStreamWrapper> stream, Ptr<NetDevice> nd);will call the device implementation of EnableAsciiInternal directly, providing either a valid prefix or a stream. All the other public ASCII tracing methods build on these low-level functions to hand you extra user-level functionality. What that means for you: all device helpers in the system get all the ASCII trace methods, and they'll all work the same way across devices — as long as the devices implement EnableAsciiInternal correctly. 🦋
✨ Methods ✨
void EnableAscii (std::string prefix, Ptr<NetDevice> nd, bool explicitFilename = false);
void EnableAscii (Ptr<OutputStreamWrapper> stream, Ptr<NetDevice> nd);
void EnableAscii (std::string prefix, std::string ndName, bool explicitFilename = false);
void EnableAscii (Ptr<OutputStreamWrapper> stream, std::string ndName);
void EnableAscii (std::string prefix, NetDeviceContainer d);
void EnableAscii (Ptr<OutputStreamWrapper> stream, NetDeviceContainer d);
void EnableAscii (std::string prefix, NodeContainer n);
void EnableAscii (Ptr<OutputStreamWrapper> stream, NodeContainer n);
void EnableAsciiAll (std::string prefix);
void EnableAsciiAll (Ptr<OutputStreamWrapper> stream);
void EnableAscii (std::string prefix, uint32_t nodeid, uint32_t deviceid, bool explicitFilename);
void EnableAscii (Ptr<OutputStreamWrapper> stream, uint32_t nodeid, uint32_t deviceid);You're encouraged to peruse the API Documentation for class AsciiTraceHelperForDevice for the full details, but to summarize …
- There are twice as many methods available for ASCII tracing as there were for PCAP tracing. Why? Because on top of the PCAP-style model — where traces from each unique node/device pair go to their own unique file — we also support a model where trace info for many node/device pairs gets written to a common file. So the
<prefix>-<node>-<device>filename generation mechanism gets replaced by a mechanism that refers to a common file, and the number of API methods doubles so all combinations are covered. 💅 - Just like in PCAP tracing, you can enable ASCII tracing on a particular (node, net-device) pair by handing a
Ptr<NetDevice>to anEnableAsciimethod. ThePtr<Node>is implicit since the net device must belong to exactly one Node. Example:
Ptr<NetDevice> nd;
...
helper.EnableAscii ("prefix", nd);- The first four methods also include a default parameter called
explicitFilename, which operates similarly to the equivalent parameter in the PCAP case.
In this case, no trace contexts get written to the ASCII trace file, since they'd just be redundant. The system picks the filename to create using the same rules as described in the PCAP section — except the file gets the suffix .tr instead of .pcap. 🎀
- If you wanna enable ASCII tracing on more than one net device and have all the traces sent to a single file, you can do that too by using an object to refer to a single file. We've already seen this in the
cwndexample above:
Ptr<NetDevice> nd1;
Ptr<NetDevice> nd2;
...
Ptr<OutputStreamWrapper> stream = asciiTraceHelper.CreateFileStream ("trace-file-name.tr");
...
helper.EnableAscii (stream, nd1);
helper.EnableAscii (stream, nd2);- In this case, trace contexts get written straight into the ASCII trace file, because bestie we need them to tell the two devices' traces apart. And heads up: since you're fully specifying the filename here, your string should include the
,trsuffix for consistency. periodt. - You can turn on ASCII tracing for one specific (node, net-device) pair by handing an
std::stringthat's an object name service string to anEnablePcapmethod. ThePtr<NetDevice>gets looked up from that name string. And again, the<Node>is implicit because a named net device has to belong to exactly one Node. Like:
Names::Add ("client" ...);
Names::Add ("client/eth0" ...);
Names::Add ("server" ...);
Names::Add ("server/eth0" ...);
...
helper.EnableAscii ("prefix", "client/eth0");
helper.EnableAscii ("prefix", "server/eth0");This gives you two files, prefix-client-eth0.tr and prefix-server-eth0.tr, with traces for each device living in its own respective trace file. And since all of the EnableAscii functions are overloaded to take a stream wrapper, you can use that form too:
Names::Add ("client" ...);
Names::Add ("client/eth0" ...);
Names::Add ("server" ...);
Names::Add ("server/eth0" ...);
...
Ptr<OutputStreamWrapper> stream = asciiTraceHelper.CreateFileStream ("trace-file-name.tr");
...
helper.EnableAscii (stream, "client/eth0");
helper.EnableAscii (stream, "server/eth0");That one gives you a single trace file called trace-file-name.tr holding every trace event for both devices, and the events get disambiguated by their trace context strings. 💖
- You can also enable ASCII tracing on a whole collection of (node, net-device) pairs by passing a
NetDeviceContainer. For everyNetDevicein the container the type gets checked, and for each device of the proper type — the same type the device helper manages — tracing gets enabled. Once again the<Node>is implicit, since a found net device must belong to exactly one Node. Example:
NetDeviceContainer d = ...;
...
helper.EnableAscii ("prefix", d);This creates a bunch of ASCII trace files, each one following the <prefix>-<node id>-<device id>.tr convention.
Combining all those traces into one single file works just like the examples above:
NetDeviceContainer d = ...;
...
Ptr<OutputStreamWrapper> stream = asciiTraceHelper.CreateFileStream ("trace-file-name.tr");
...
helper.EnableAscii (stream, d);- You can enable ASCII tracing on a collection of (node, net-device) pairs by passing a
NodeContainertoo. For each Node in the NodeContainer its attached NetDevices get iterated, and for each NetDevice attached to each Node in the container, that device's type gets checked. For each device of the proper type (same type as the device helper manages), tracing is enabled.
NodeContainer n;
...
helper.EnableAscii ("prefix", n);Same deal: you get a bunch of ASCII trace files, each following the <prefix>-<node id>-<device id>.tr convention, and combining them all into a single file works just like the examples above.
- You can enable ASCII tracing based on Node ID and device ID as well as with an explicit
Ptr. Every Node in the system has an integer Node ID, and every device connected to a Node has an integer device ID.
helper.EnableAscii ("prefix", 21, 1);And of course, the traces can be combined into a single file as shown above.
- Finally, you can enable ASCII tracing for every single device in the system, as long as it's the same type the device helper manages.
helper.EnableAsciiAll ("prefix");This creates a bunch of ASCII trace files — one for every device in the system of the type managed by the helper. All of these files follow the <prefix>-<node id>-<device id>.tr convention. And yes, combining all of the traces into a single file is accomplished similarly to the examples above. 🦋
🎀 Filenames 🎀
Lowkey the whole prefix-style method thing above has the construction of the complete filenames baked into the implementation method. By convention, ASCII traces in the ns-3 system look like <prefix>-<node id>-<device id>.tr
Like we said earlier, every Node in the system has a system-assigned Node id, and every device has an interface index (also called a device id) relative to its node. So by default, an ASCII trace file created by enabling tracing on the first device of Node 21 using the prefix "prefix" would be prefix-21-1.tr.
You can always use the ns-3 object name service to make this clearer. For example, if you use the object name service to assign the name "server" to Node 21, the resulting ASCII trace file name automatically becomes prefix-server-1.tr — and if you also assign the name "eth0" to the device, your ASCII trace file name automatically picks that up too and gets called prefix-server-eth0.tr. iykyk ✨
Several of the methods have a default parameter called explicitFilename. When it's set to true, that parameter disables the automatic filename completion mechanism and lets you create an explicit filename. Heads up though: this option is only available in the methods that take a prefix and enable tracing on a single device. 💅
🌈 Protocol Helpers 🌈
💀 PCAP 💀
The goal of these mixins is to make it easy to add a consistent PCAP trace facility to protocols. We want all the various flavors of PCAP tracing to work the same across all protocols, so the methods of these helpers get inherited by stack helpers. If you wanna follow the discussion while looking at real code, go peek at src/network/helper/trace-helper.h.
In this section we're illustrating the methods as applied to the protocol Ipv4. To specify traces in similar protocols, just swap in the appropriate type. For example, use a Ptr<Ipv6> instead of a Ptr<Ipv4> and call EnablePcapIpv6 instead of EnablePcapIpv4.
The class PcapHelperForIpv4 provides the high level functionality for using PCAP tracing in the Ipv4 protocol. Each protocol helper enabling these methods must implement a single virtual method inherited from this class. There will be a separate implementation for Ipv6, for example, but the only difference will be in the method names and signatures. Different method names are required to disambiguate class Ipv4 from Ipv6 — both of which are derived from class Object — and methods that share the same signature.
virtual void EnablePcapIpv4Internal (std::string prefix,
Ptr<Ipv4> ipv4,
uint32_t interface,
bool explicitFilename) = 0;The signature of this method reflects the protocol- and interface-centric view of the situation at this level. All the public methods inherited from class PcapHelperForIpv4 reduce down to calling this single device-dependent implementation method. For example, the lowest level PCAP method,
void EnablePcapIpv4 (std::string prefix, Ptr<Ipv4> ipv4, uint32_t interface, bool explicitFilename = false);will call the device implementation of EnablePcapIpv4Internal directly. All other public PCAP tracing methods build on this implementation to provide extra user-level functionality. What that means for you, bestie, is that all protocol helpers in the system will have all of the PCAP trace methods available, and these methods will all work the same way across protocols — as long as the helper implements EnablePcapIpv4Internal correctly. 🫶
⭐ Methods ⭐
These methods are designed to be in one-to-one correspondence with the Node- and NetDevice-centric versions of the device versions. Instead of Node and NetDevice pair constraints, we use protocol and interface constraints.
And ngl, just like in the device version, there are six methods:
void EnablePcapIpv4 (std::string prefix, Ptr<Ipv4> ipv4, uint32_t interface, bool explicitFilename = false);
void EnablePcapIpv4 (std::string prefix, std::string ipv4Name, uint32_t interface, bool explicitFilename = false);
void EnablePcapIpv4 (std::string prefix, Ipv4InterfaceContainer c);
void EnablePcapIpv4 (std::string prefix, NodeContainer n);
void EnablePcapIpv4 (std::string prefix, uint32_t nodeid, uint32_t interface, bool explicitFilename);
void EnablePcapIpv4All (std::string prefix);You're encouraged to peruse the API Documentation for class PcapHelperForIpv4 to find the details of these methods, but to summarize …
- You can enable PCAP tracing on a particular protocol/interface pair by providing a
Ptr<Ipv4>and interface to anEnablePcapmethod. Like:
Ptr<Ipv4> ipv4 = node->GetObject<Ipv4> ();
...
helper.EnablePcapIpv4 ("prefix", ipv4, 0);- You can enable PCAP tracing on a particular node/net-device pair by providing a
std::stringrepresenting an object name service string to anEnablePcapmethod. ThePtr<Ipv4>gets looked up from that name string. Like:
Names::Add ("serverIPv4" ...);
...
helper.EnablePcapIpv4 ("prefix", "serverIpv4", 1);- You can enable PCAP tracing on a collection of protocol/interface pairs by providing an
Ipv4InterfaceContainer. For eachIpv4/ interface pair in the container, the protocol type gets checked. For each protocol of the proper type (the same type the device helper manages), tracing gets enabled for the corresponding interface. Like:
NodeContainer nodes;
...
NetDeviceContainer devices = deviceHelper.Install (nodes);
...
Ipv4AddressHelper ipv4;
ipv4.SetBase ("10.1.1.0", "255.255.255.0");
Ipv4InterfaceContainer interfaces = ipv4.Assign (devices);
...
helper.EnablePcapIpv4 ("prefix", interfaces);- You can enable PCAP tracing on a collection of protocol/interface pairs by providing a
NodeContainer. For each Node in the NodeContainer the appropriate protocol is found, and for each protocol its interfaces get enumerated and tracing gets enabled on the resulting pairs. Like:
NodeContainer n;
...
helper.EnablePcapIpv4 ("prefix", n);- You can enable PCAP tracing on the basis of Node ID and interface as well. In this case, the node-id gets translated to a
Ptr<Node>and the appropriate protocol gets looked up in the node. The resulting protocol and interface are then used to specify the resulting trace source.
helper.EnablePcapIpv4 ("prefix", 21, 1);- Finally, you can turn on PCAP tracing for every interface in the whole system, with the associated protocol being the same type as the one managed by the device helper. 💫
helper.EnablePcapIpv4All ("prefix");🌸 Filenames 🌸
Okay so here's the tea on how files get named. Implicit in all of the method descriptions above is the construction of the complete filenames by the implementation method. By convention, PCAP traces taken for devices in the ns-3 system are of the form "<prefix>-<node id>-<device id>.pcap". ✨
For protocol traces though, it's a little different. There is a one-to-one correspondence between protocols and Nodes, and that's because protocol Objects are aggregated to Node Objects. Since there's no global protocol id in the system, we use the corresponding Node id in file naming. Which means… there is a possibility for file name collisions in automatically chosen trace file names. 😬 So for that reason, the file name convention is changed for protocol traces.
As mentioned before, every Node in the system has a system-assigned Node id. Since there's a one-to-one correspondence between protocol instances and Node instances, we use the Node id. Each interface has an interface id relative to its protocol. The convention here is "<prefix>-n<node id>-i<interface id>.pcap" for trace file naming in protocol helpers.
So by default, a PCAP trace file created from enabling tracing on interface 1 of the Ipv4 protocol of Node 21, using the prefix "prefix", would be "prefix-n21-i1.pcap". 💅
You can always use the ns-3 object name service to make this clearer. For example, if you use the object name service to assign the name "serverIpv4" to the Ptr<Ipv4> on Node 21, the resulting PCAP trace file name automatically becomes "prefix-nserverIpv4-i1.pcap". 🎀
Several of the methods have a default parameter called explicitFilename. When it's set to true, this parameter disables the automatic filename completion mechanism and lets you create an explicit filename. Heads up though — this option is only available in the methods which take a prefix and enable tracing on a single device. 🫶
🦋 ASCII 🦋
The behavior of the ASCII trace helpers is substantially similar to the PCAP case. Take a look at src/network/helper/trace-helper.h if you want to follow along while looking at real code. 📖
In this section we'll be illustrating the methods as applied to the protocol Ipv4. To specify traces in similar protocols, just swap in the appropriate type. For example, use a Ptr<Ipv6> instead of a Ptr<Ipv4> and call EnableAsciiIpv6 instead of EnableAsciiIpv4. ✨
The class AsciiTraceHelperForIpv4 adds the high level functionality for using ASCII tracing to a protocol helper. Each protocol that enables these methods must implement a single virtual method inherited from this class. 👑
virtual void EnableAsciiIpv4Internal (Ptr<OutputStreamWrapper> stream,
std::string prefix,
Ptr<Ipv4> ipv4,
uint32_t interface,
bool explicitFilename) = 0;The signature of this method reflects the protocol- and interface-centric view of the situation at this level, and also the fact that the helper may be writing to a shared output stream. All of the public methods inherited from class PcapAndAsciiTraceHelperForIpv4 reduce to calling this single device-dependent implementation method. For example, the lowest level ASCII trace methods:
void EnableAsciiIpv4 (std::string prefix, Ptr<Ipv4> ipv4, uint32_t interface, bool explicitFilename = false);
void EnableAsciiIpv4 (Ptr<OutputStreamWrapper> stream, Ptr<Ipv4> ipv4, uint32_t interface);will call the device implementation of EnableAsciiIpv4Internal directly, providing either the prefix or the stream. All other public ASCII tracing methods build on these low-level functions to provide extra user-level functionality. What this means for you is that all device helpers in the system will have all of the ASCII trace methods available, and these methods will all work the same way across protocols — if the protocols implement EnableAsciiIpv4Internal correctly. no cap. 💯
💜 Methods 💜
void EnableAsciiIpv4 (std::string prefix, Ptr<Ipv4> ipv4, uint32_t interface, bool explicitFilename = false);
void EnableAsciiIpv4 (Ptr<OutputStreamWrapper> stream, Ptr<Ipv4> ipv4, uint32_t interface);
void EnableAsciiIpv4 (std::string prefix, std::string ipv4Name, uint32_t interface, bool explicitFilename = false);
void EnableAsciiIpv4 (Ptr<OutputStreamWrapper> stream, std::string ipv4Name, uint32_t interface);
void EnableAsciiIpv4 (std::string prefix, Ipv4InterfaceContainer c);
void EnableAsciiIpv4 (Ptr<OutputStreamWrapper> stream, Ipv4InterfaceContainer c);
void EnableAsciiIpv4 (std::string prefix, NodeContainer n);
void EnableAsciiIpv4 (Ptr<OutputStreamWrapper> stream, NodeContainer n);
void EnableAsciiIpv4All (std::string prefix);
void EnableAsciiIpv4All (Ptr<OutputStreamWrapper> stream);
void EnableAsciiIpv4 (std::string prefix, uint32_t nodeid, uint32_t deviceid, bool explicitFilename);
void EnableAsciiIpv4 (Ptr<OutputStreamWrapper> stream, uint32_t nodeid, uint32_t interface);You are encouraged to peruse the API Documentation for class PcapAndAsciiHelperForIpv4 to find the details of these methods, but to summarize… 👇
- There are twice as many methods available for ASCII tracing as there were for PCAP tracing. This is because, in addition to the PCAP-style model where traces from each unique protocol/interface pair are written to a unique file, we support a model in which trace information for many protocol/interface pairs is written to a common file. This means that the
<prefix>-n<node id>-<interface>file name generation mechanism is replaced by a mechanism to refer to a common file, and the number of API methods is doubled to allow all the combinations. 🎀 - Just like in PCAP tracing, you can enable ASCII tracing on a particular protocol/interface pair by providing a
Ptr<Ipv4>and an interface to anEnableAsciimethod. For example:
Ptr<Ipv4> ipv4;
...
helper.EnableAsciiIpv4 ("prefix", ipv4, 1);In this case, no trace contexts are written to the ASCII trace file since they would be redundant. The system will pick the file name to be created using the same rules as described in the PCAP section, except that the file will have the suffix ".tr" instead of ".pcap". ✨
- If you want to enable ASCII tracing on more than one interface and have all traces sent to a single file, you can do that too by using an object to refer to a single file. We already have something similar to this in the
"cwnd"example above:
Ptr<Ipv4> protocol1 = node1->GetObject<Ipv4> ();
Ptr<Ipv4> protocol2 = node2->GetObject<Ipv4> ();
...
Ptr<OutputStreamWrapper> stream = asciiTraceHelper.CreateFileStream ("trace-file-name.tr");
...
helper.EnableAsciiIpv4 (stream, protocol1, 1);
helper.EnableAsciiIpv4 (stream, protocol2, 1);In this case, trace contexts are written to the ASCII trace file since they're required to disambiguate traces from the two interfaces. Note that since the user is completely specifying the file name, the string should include the ",tr" for consistency. 💅
- You can enable ASCII tracing on a particular protocol by providing a
std::stringrepresenting an object name service string to anEnablePcapmethod. ThePtr<Ipv4>is looked up from the name string. The<Node>in the resulting filenames is implicit since there's a one-to-one correspondence between protocol instances and nodes. For example:
Names::Add ("node1Ipv4" ...);
Names::Add ("node2Ipv4" ...);
...
helper.EnableAsciiIpv4 ("prefix", "node1Ipv4", 1);
helper.EnableAsciiIpv4 ("prefix", "node2Ipv4", 1);This would result in two files named "prefix-nnode1Ipv4-i1.tr" and "prefix-nnode2Ipv4-i1.tr", with traces for each interface in the respective trace file. Since all of the EnableAscii functions are overloaded to take a stream wrapper, you can use that form as well:
Names::Add ("node1Ipv4" ...);
Names::Add ("node2Ipv4" ...);
...
Ptr<OutputStreamWrapper> stream = asciiTraceHelper.CreateFileStream ("trace-file-name.tr");
...
helper.EnableAsciiIpv4 (stream, "node1Ipv4", 1);
helper.EnableAsciiIpv4 (stream, "node2Ipv4", 1);This would result in a single trace file called "trace-file-name.tr" that contains all of the trace events for both interfaces. The events would be disambiguated by trace context strings. 🌈
- You can enable ASCII tracing on a collection of protocol/interface pairs by providing an
Ipv4InterfaceContainer. For each protocol of the proper type (the same type as is managed by the device helper), tracing is enabled for the corresponding interface. Again, the<Node>is implicit since there's a one-to-one correspondence between each protocol and its node. For example:
NodeContainer nodes;
...
NetDeviceContainer devices = deviceHelper.Install (nodes);
...
Ipv4AddressHelper ipv4;
ipv4.SetBase ("10.1.1.0", "255.255.255.0");
Ipv4InterfaceContainer interfaces = ipv4.Assign (devices);
...
...
helper.EnableAsciiIpv4 ("prefix", interfaces);This would result in a number of ASCII trace files being created, each of which follows the <prefix>-n<node id>-i<interface>.tr convention. Combining all of the traces into a single file is accomplished similarly to the examples above:
NodeContainer nodes;
...
NetDeviceContainer devices = deviceHelper.Install (nodes);
...
Ipv4AddressHelper ipv4;
ipv4.SetBase ("10.1.1.0", "255.255.255.0");
Ipv4InterfaceContainer interfaces = ipv4.Assign (devices);
...
Ptr<OutputStreamWrapper> stream = asciiTraceHelper.CreateFileStream ("trace-file-name.tr");
...
helper.EnableAsciiIpv4 (stream, interfaces);- You can enable ASCII tracing on a collection of protocol/interface pairs by providing a
NodeContainer. For each Node in theNodeContainerthe appropriate protocol is found. For each protocol, its interfaces are enumerated and tracing is enabled on the resulting pairs. For example:
NodeContainer n;
...
helper.EnableAsciiIpv4 ("prefix", n);This would result in a number of ASCII trace files being created, each of which follows the <prefix>-<node id>-<device id>.tr convention. Combining all of the traces into a single file is accomplished similarly to the examples above. 🫶
- You can flip ASCII tracing on by Node ID and device ID too, bestie. In that case the node-id gets translated to a
Ptr<Node>and the right protocol gets looked up inside the node. Then that resulting protocol + interface pair is what specifies the trace source you end up with. It's giving "precision targeting" 💅
helper.EnableAsciiIpv4 ("prefix", 21, 1);And yeah, of course you can smash all those traces into a single file, same as the example above. ✨
- And FINALLY, the "I want literally everything" option: you can enable ASCII tracing for all interfaces in the system, with the associated protocol being the same type as the one managed by the device helper.
helper.EnableAsciiIpv4All ("prefix");This creates a whole lil army of ASCII trace files, one for every interface in the system related to a protocol of the type managed by the helper. All of them follow the <prefix>-n<node id>-i<interface.tr convention. 🦋 And combining them all into one single file? Same deal as the examples above.
🎀 Filenames 🎀
So the whole prefix-style thing we've been talking about is lowkey hiding something: the implementation method is the one constructing the complete filenames. By convention, ASCII traces in the ns-3 system look like:
“<prefix>-<node id>-<device id>.tr”
As we said before, every Node in the system has a system-assigned Node id. And because there's a one-to-one correspondence between protocols and nodes, we use the node-id to identify the protocol identity. 🌸 Every interface on a given protocol has an interface index (also just called an interface) relative to its protocol.
So by default, an ASCII trace file created by enabling tracing on the first device of Node 21, using the prefix “prefix”, would be “prefix-n21-i1.tr”. Use the prefix to disambiguate multiple protocols per node — that's its whole job fr fr. 💜
And you can always use the ns-3 object name service to make this clearer. For example, if you use the object name service to assign the name “serverIpv4” to the protocol on Node 21, and you also specify interface one, the resulting ASCII trace file name automatically becomes “prefix-nserverIpv4-1.tr”. Iconic. 👑
Several of the methods have a default parameter called explicitFilename. When you set it to true, it disables the automatic filename completion mechanism and lets you create an explicit filename. 🫶 Heads up though: this option is only available in the methods which take a prefix and enable tracing on a single device.
✨ Summary ✨
ns-3 includes an extremely rich environment allowing users at several levels to customize the kinds of information that can be extracted from simulations. It's a lot, bestie, but it's a lot in a good way. 💖
- There are high-level helper functions that let users simply control the collection of pre-defined outputs down to a fine granularity.
- There are mid-level helper functions to let more sophisticated users customize how information is extracted and saved.
- And there are low-level core functions</low-level> to let expert users alter the system so it presents new and previously unexported information in a way that'll be immediately accessible to users at higher levels. ⭐
This is a very comprehensive system, and we realize it is a lot to digest — especially for new users, or those not intimately familiar with C++ and its idioms. 😭 We do consider the tracing system a very important part of ns-3, so we recommend becoming as familiar as possible with it. And honestly? It is probably the case that understanding the rest of the ns-3 system will be quite simple once you have mastered the tracing system. 🧚
💖 DATA COLLECTION 💖
Our final tutorial chapter introduces some components that were added to ns-3 in version 3.18, and that are still under development. This tutorial section is also a work-in-progress. 🙈 No cap, we're building the plane while flying it.
🌸 Motivation 🌸
One of the main points of running simulations is to generate output data — either for research purposes or simply to learn about the system. In the previous chapter, we introduced the tracing subsystem and the example sixth.cc. from which PCAP or ASCII trace files are generated. Those traces are valuable for data analysis using a variety of external tools, and for many users, that kind of output data is a preferred means of gathering data (for analysis by external tools). 💅
But there are also use cases for more than trace file generation, including:
- generation of data that does not map well to PCAP or ASCII traces, such as non-packet data (e.g. protocol state machine transitions),
- large simulations for which the disk I/O requirements for generating trace files is prohibitive or cumbersome, and
- the need for online data reduction or computation, during the course of the simulation. A good example of this is to define a termination condition for the simulation, to tell it when to stop when it has received enough data to form a narrow-enough confidence interval around the estimate of some parameter. 🔥
The ns-3 data collection framework is designed to provide these additional capabilities beyond trace-based output. We recommend that the reader interested in this topic consult the ns-3 Manual for a more detailed treatment of this framework; here, we summarize with an example program some of the developing capabilities. 🦋
🌈 Example Code 🌈
The tutorial example examples/tutorial/seventh.cc resembles the sixth.cc example we previously reviewed, except for a few changes. First, it has been enabled for IPv6 support with a command-line option:
CommandLine cmd;
cmd.AddValue ("useIpv6", "Use Ipv6", useV6);
cmd.Parse (argc, argv);If the user specifies useIpv6, option, the program will be run using IPv6 instead of IPv4. The help option — available on all ns-3 programs that support the CommandLine object as shown above — can be invoked as follows (please note the use of double quotes):
./waf --run "seventh --help"which produces:
ns3-dev-seventh-debug [Program Arguments] [General Arguments]
Program Arguments:
--useIpv6: Use Ipv6 [false]
General Arguments:
--PrintGlobals: Print the list of globals.
--PrintGroups: Print the list of groups.
--PrintGroup=[group]: Print all TypeIds of group.
--PrintTypeIds: Print all TypeIds.
--PrintAttributes=[typeid]: Print all attributes of typeid.
--PrintHelp: Print this help message.This default (use of IPv4, since useIpv6 is false) can be changed by toggling the boolean value as follows:
./waf --run "seventh --useIpv6=1"and have a look at the pcap generated, such as with tcpdump:
tcpdump -r seventh.pcap -nn -ttThis has been a short digression into IPv6 support and the command line, which was also introduced earlier in this tutorial. For a dedicated example of command line usage, please see src/core/examples/command-line-example.cc. 💜
Now back to data collection. In the examples/tutorial/ directory, type the following command: diff -u sixth.cc seventh.cc, and examine some of the new lines of this diff:
+ std::string probeType;
+ std::string tracePath;
+ if (useV6 == false)
+ {
...
+ probeType = "ns3::Ipv4PacketProbe";
+ tracePath = "/NodeList/*/$ns3::Ipv4L3Protocol/Tx";
+ }
+ else
+ {
...
+ probeType = "ns3::Ipv6PacketProbe";
+ tracePath = "/NodeList/*/$ns3::Ipv6L3Protocol/Tx";
+ }
...
+ // Use GnuplotHelper to plot the packet byte count over time
+ GnuplotHelper plotHelper;
+
+ // Configure the plot. The first argument is the file name prefix
+ // for the output files generated. The second, third, and fourth
+ // arguments are, respectively, the plot title, x-axis, and y-axis labels
+ plotHelper.ConfigurePlot ("seventh-packet-byte-count",
+ "Packet Byte Count vs. Time",
+ "Time (Seconds)",
+ "Packet Byte Count");
+
+ // Specify the probe type, trace source path (in configuration namespace), and
+ // probe output trace source ("OutputBytes") to plot. The fourth argument
+ // specifies the name of the data series label on the plot. The last
+ // argument formats the plot by specifying where the key should be placed.
+ plotHelper.PlotProbe (probeType,
+ tracePath,
+ "OutputBytes",
+ "Packet Byte Count",
+ GnuplotAggregator::KEY_BELOW);
+
+ // Use FileHelper to write out the packet byte count over time
+ FileHelper fileHelper;
+
+ // Configure the file to be written, and the formatting of output data.
+ fileHelper.ConfigureFile ("seventh-packet-byte-count",
+ FileAggregator::FORMATTED);
+
+ // Set the labels for this formatted output file.
+ fileHelper.Set2dFormat ("Time (Seconds) = %.3e\tPacket Byte Count = %.0f");
+
+ // Specify the probe type, probe path (in configuration namespace), and
+ // probe output trace source ("OutputBytes") to write.
+ fileHelper.WriteProbe (probeType,
+ tracePath,
+ "OutputBytes");
+
Simulator::Stop (Seconds (20));
Simulator::Run ();
Simulator::Destroy ();The careful reader will have noticed, when testing the IPv6 command line attribute above, that seventh.cc had created a number of new output files:
seventh-packet-byte-count-0.txt
seventh-packet-byte-count-1.txt
seventh-packet-byte-count.dat
seventh-packet-byte-count.plt
seventh-packet-byte-count.png
seventh-packet-byte-count.shThese were created by the additional statements introduced above — in particular, by a GnuplotHelper and a FileHelper. 🎀 This data was produced by hooking the data collection components to ns-3 trace sources, and marshaling the data into a formatted gnuplot and into a formatted text file. In the next sections, we'll review each of these. ✨
👑 GnuplotHelper 👑
The GnuplotHelper is an ns-3 helper object aimed at the production of gnuplot plots with as few statements as possible, for common cases. It hooks ns-3 trace sources with data types supported by the data collection system. Not all ns-3 trace sources data types are supported, but many of the common trace types are, including TracedValues with plain old data (POD) types. 💅
Let's look at the output produced by this helper:
seventh-packet-byte-count.dat
seventh-packet-byte-count.plt
seventh-packet-byte-count.shokay bestie, so the first file is a gnuplot data file — like, a whole series of space-delimited timestamps plus packet byte counts. We're gonna cover how this specific data output got configured down below, but bestie, let's keep going with the output files first, we're on a roll 💅
The file seventh-packet-byte-count.plt is a gnuplot plot file, and you can open it from inside gnuplot. If you already speak gnuplot syntax, you'll see it produces a formatted output PNG file named seventh-packet-byte-count.png. And finally? A tiny little shell script, seventh-packet-byte-count.sh, runs this plot file through gnuplot to make the PNG you actually wanted (which you can open in an image editor). So basically, this command:
sh seventh-packet-byte-count.shwill hand you seventh-packet-byte-count.png. ✨
Now, the question every icon asks: why wasn't this PNG just produced in the first place?? 💀 The answer is literally that by giving you the plt file, you can hand-configure the result yourself if you want, before the PNG gets produced. Customization, bestie. We love options.
The PNG image title states that this plot is a plot of "Packet Byte Count vs. Time", and that it's plotting the probed data corresponding to the trace source path:
/NodeList/*/$ns3::Ipv6L3Protocol/TxNote the wild-card in the trace path — yes, the * is doing the heavy lifting there, main character energy. In summary, what this plot is capturing is the plot of packet bytes observed at the transmit trace source of the Ipv6L3Protocol object; largely 596-byte TCP segments in one direction, and 60-byte TCP acks in the other (two node trace sources were matched by this trace source, no cap).
💖 How Was This Configured? 💖
A few statements need to be provided. First, the GnuplotHelper object must be declared and configured:
+ // Use GnuplotHelper to plot the packet byte count over time
+ GnuplotHelper plotHelper;
+
+ // Configure the plot. The first argument is the file name prefix
+ // for the output files generated. The second, third, and fourth
+ // arguments are, respectively, the plot title, x-axis, and y-axis labels
+ plotHelper.ConfigurePlot ("seventh-packet-byte-count",
+ "Packet Byte Count vs. Time",
+ "Time (Seconds)",
+ "Packet Byte Count");So at this point, an empty plot has been configured, bestie. The filename prefix is the first argument, the plot title is the second, the x-axis label is the third, and the y-axis label is the fourth argument. Simple as that, fr fr.
Next step is configuring the data, and this is where the trace source gets hooked 🎣. First, note that above in the program we declared a few variables for later use:
+ std::string probeType;
+ std::string tracePath;
+ probeType = "ns3::Ipv6PacketProbe";
+ tracePath = "/NodeList/*/$ns3::Ipv6L3Protocol/Tx";And we use them right here:
+ // Specify the probe type, trace source path (in configuration namespace), and
+ // probe output trace source ("OutputBytes") to plot. The fourth argument
+ // specifies the name of the data series label on the plot. The last
+ // argument formats the plot by specifying where the key should be placed.
+ plotHelper.PlotProbe (probeType,
+ tracePath,
+ "OutputBytes",
+ "Packet Byte Count",
+ GnuplotAggregator::KEY_BELOW);The first two arguments are the name of the probe type and the trace source path. And ngl bestie, these two are probably the hardest to figure out when you're trying to use this framework to plot other traces 😭. The probe trace here is the Tx trace source of class Ipv6L3Protocol. When we examine this class implementation (src/internet/model/ipv6-l3-protocol.cc) we can observe:
.AddTraceSource ("Tx", "Send IPv6 packet to outgoing interface.",
MakeTraceSourceAccessor (&Ipv6L3Protocol::m_txTrace))This says that Tx is a name for variable m_txTrace, which has a declaration of:
/**
* \brief Callback to trace TX (transmission) packets.
*/
TracedCallback<Ptr<const Packet>, Ptr<Ipv6>, uint32_t> m_txTrace;It turns out that this specific trace source signature is supported by a Probe class (which is what we need here) of class Ipv6PacketProbe. Go look at the files src/internet/model/ipv6-packet-probe.{h,cc} — iconic, honestly. 🦋
So, in the PlotProbe statement above, we see that the statement is hooking the trace source (identified by path string) with a matching ns-3 Probe type of Ipv6PacketProbe. And if we did not support this probe type (matching trace source signature), we could have not used this statement — although some more complicated lower-level statements could have been used, as described in the manual. So like, options exist, but they're giving extra credit energy.
The Ipv6PacketProbe exports, itself, some trace sources that extract the data out of the probed Packet object:
TypeId
Ipv6PacketProbe::GetTypeId ()
{
static TypeId tid = TypeId ("ns3::Ipv6PacketProbe")
.SetParent<Probe> ()
.SetGroupName ("Stats")
.AddConstructor<Ipv6PacketProbe> ()
.AddTraceSource ( "Output",
"The packet plus its IPv6 object and interface that serve as the output for this probe",
MakeTraceSourceAccessor (&Ipv6PacketProbe::m_output))
.AddTraceSource ( "OutputBytes",
"The number of bytes in the packet",
MakeTraceSourceAccessor (&Ipv6PacketProbe::m_outputBytes))
;
return tid;
}The third argument of our PlotProbe statement specifies that we're interested in the number of bytes in this packet; specifically, the "OutputBytes" trace source of Ipv6PacketProbe. Finally, the last two arguments of the statement provide the plot legend for this data series ("Packet Byte Count"), and an optional gnuplot formatting statement (GnuplotAggregator::KEY_BELOW) that we want the plot key to be inserted below the plot. Other options include NO_KEY, KEY_INSIDE, and KEY_ABOVE. Pick your vibe ✨
🌸 Supported Trace Types 🌸
The following traced values are supported with Probes as of this writing:
double—DoubleProbe—stats/model/double-probe.huint8_t—Uinteger8Probe—stats/model/uinteger-8-probe.huint16_t—Uinteger16Probe—stats/model/uinteger-16-probe.huint32_t—Uinteger32Probe—stats/model/uinteger-32-probe.hbool—BooleanProbe—stats/model/uinteger-16-probe.hns3::Time—TimeProbe—stats/model/time-probe.h
The following TraceSource types are supported by Probes as of this writing:
Ptr<const Packet>—PacketProbe— probe outputsOutputBytes—network/utils/packet-probe.hPtr<const Packet>, Ptr<Ipv4>, uint32_t—Ipv4PacketProbe— probe outputsOutputBytes—internet/model/ipv4-packet-probe.hPtr<const Packet>, Ptr<Ipv6>, uint32_t—Ipv6PacketProbe— probe outputsOutputBytes—internet/model/ipv6-packet-probe.hPtr<const Packet>, Ptr<Ipv6>, uint32_t—Ipv6PacketProbe— probe outputsOutputBytes—internet/model/ipv6-packet-probe.hPtr<const Packet>, const Address&—ApplicationPacketProbe— probe outputsOutputBytes—applications/model/application-packet-probe.h
Okay so, real talk: only a few trace sources are supported here, and every single one of them is oriented toward outputting the Packet size (in bytes). That's it. That's the whole buffet. But bestie, most of the fundamental data types available as TracedValues can be supported with these helpers — so it's giving "starter pack", not "full closet". ✨
🌸 FileHelper 🌸
The FileHelper class is just a little variation on the GnuplotHelper example we already did. The example program gives you formatted output of the same timestamped data, like this:
Time (Seconds) = 9.312e+00 Packet Byte Count = 596
Time (Seconds) = 9.312e+00 Packet Byte Count = 564Two files get written — one for node "0" and one for node "1", and you can literally see that in the filenames. iykyk. Let's go through the code piece-by-piece 💅
+ // Use FileHelper to write out the packet byte count over time
+ FileHelper fileHelper;
+
+ // Configure the file to be written, and the formatting of output data.
+ fileHelper.ConfigureFile ("seventh-packet-byte-count",
+ FileAggregator::FORMATTED);So the file helper file prefix is the first argument, and the format specifier comes next. Other formatting options include SPACE_SEPARATED, COMMA_SEPARATED, and TAB_SEPARATED. And if you went with FORMATTED, you're allowed to change the formatting with a format string, like this:
+
+ // Set the labels for this formatted output file.
+ fileHelper.Set2dFormat ("Time (Seconds) = %.3e\tPacket Byte Count = %.0f");Finally — and this is the main character moment — the trace source of interest has to be hooked. Again, the probeType and tracePath variables in this example are used, and the probe's output trace source "OutputBytes" is what gets hooked:
+
+ // Specify the probe type, trace source path (in configuration namespace), and
+ // probe output trace source ("OutputBytes") to write.
+ fileHelper.WriteProbe (probeType,
+ tracePath,
+ "OutputBytes");
+The wildcard fields in this trace source specifier match two trace sources. And here's the plot twist: unlike the GnuplotHelper example, where two data series got overlaid on the same plot — here, two separate files get written to disk. Different files, same energy. 🦋
💫 Summary 💫
Data collection support is new as of ns-3.18, and basic support for providing time series output has been added. The basic pattern described above may be replicated within the scope of support of the existing probes and trace sources. More capabilities — including statistics processing — will be added in future releases. So watch this space, bestie. 👑
🎀 CONCLUSION 🎀
🌈 Futures 🌈
This document is intended as a living document. We hope and expect it to grow over time to cover more and more of the nuts and bolts of ns-3.
Writing manual and tutorial chapters is not something we all get excited about (not me pretending to be hyped about documentation, fr fr), but it is very important to the project. If you are an expert in one of these areas, please consider contributing to ns-3 by providing one of these chapters; or any other chapter you may think is important. Be the chapter you wish to see in the world. 🫶
🌸 Closing 🌸
ns-3 is a large and complicated system. It is impossible to cover all the things you will need to know in one small tutorial. Readers who want to learn more are encouraged to read the following additional documentation:
- The ns-3 manual
- The ns-3 model library documentation
- The ns-3 Doxygen (API documentation)
- The ns-3 wiki
— The ns-3 development team. 💜
👑 AUTHOR 👑
ns-3 project
📜 COPYRIGHT 📜
2006-2019
And that's a wrap on the whole thing, bestie!! 🎉💖✨ This page came from ns-3.35, stamped with the date 1643791691 and filed under NS-3-TUTORIAL(1). Go build something iconic, run your traces, and remember: you ate. periodt. 💅🌈
📦 where this page comes from
- ubuntu jammy: ns3-doc 3.35+dfsg-1ubuntu1
- 💖 girlified by deepseek-ai/DeepSeek-V4.1-Flash on : a language model's rewrite of the original, under its terms