RFC 2681

A Round-trip Delay Metric for IPPM

📅 published September 1999, from ietf

RFC 2681: A Round-trip Delay Metric for IPPM 🎀

okay bestie, this is RFC 2681 and it's giving "how to measure round-trip delay properly" 💅 ✨

Who wrote it: G. Almes, S. Kalidindi, and M. Zekauskas, all from Advanced Network & Services 🫶 Network Working Group 💜 Category: Standards Track Date: September 1999 Obsoletes / Updates: the title block doesn't list anything it obsoletes or updates 🦋

Status of this Memo 📜

ok so this document is an Internet standards track protocol for the Internet community, and it's literally asking for discussion and suggestions for improvements 🙌 For the standardization state and status of this protocol, check the current edition of the "Internet Official Protocol Standards" (STD 1). Distribution of this memo is unlimited 🌈

Copyright (C) The Internet Society (1999). All Rights Reserved.

1. Introduction ✨

this memo defines a metric for round-trip delay of packets across Internet paths 🧚 It builds on ideas from the IPPM Framework document, RFC 2330 [1], and follows the One-way Delay metric ("A One-way Delay Metric for IPPM") [2] super closely. You're assumed to already know those documents, so no pressure 💖

the memo was mostly written by copying material from the One-way Delay metric 🍓 The plan is that where the two metrics are similar, they use the same or almost the same text, and where they differ, there's new or changed text.

this memo is meant to be structured like a future companion document for Packet Loss 🦋

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in RFC 2119 [6] 💅 RFC 2119 was written with protocols in mind, but the key words get used here for similar reasons. They make sure results from two different implementations are comparable, and they call out when an implementation could perturb the network 👑

the memo is structured like this 🌸

  • A 'singleton' analytic metric, called Type-P-Round-trip-Delay, will be introduced to measure a single observation of round-trip delay.
  • Using this singleton metric, a 'sample', called Type-P-Round-trip-Delay-Poisson-Stream, will be introduced to measure a sequence of singleton delays measured at times taken from a Poisson process.
  • Using this sample, several 'statistics' of the sample will be defined and discussed.

this singleton → sample → statistics progression, with clear separation between each step, is important. Don't skip it, bestie 🫶

whenever a technical term from the IPPM Framework document is first used in this memo, it gets a trailing asterisk. So "term*" means "term" is defined in the Framework 🧚

1.1. Motivation 💖

round-trip delay of a Type-P* packet from a source host* to a destination host is useful for a few reasons, and it's giving main character energy ✨

  • Some applications don't perform well (or at all) if end-to-end delay between hosts is large relative to some threshold value.
  • Erratic variation in delay makes it difficult (or impossible) to support many interactive real-time applications.
  • The larger the value of delay, the harder it is for transport-layer protocols to sustain high bandwidths.
  • The minimum value of this metric gives an indication of the delay due only to propagation and transmission delay.
  • The minimum value of this metric gives an indication of the delay that will likely be experienced when the path* traversed is lightly loaded.
  • Values of this metric above the minimum give an indication of the congestion present in the path.

ok but measuring round-trip delay instead of one-way delay has a few weaknesses, and we need to be honest about them 💀

  • The Internet path from a source to a destination may differ from the path from the destination back to the source ("asymmetric paths"), so different sequences of routers are used for the forward and reverse paths. That means round-trip measurements actually measure the performance of two distinct paths together 😭
  • Even when the two paths are symmetric, they may have radically different performance characteristics due to asymmetric queueing.
  • Performance of an application may depend mostly on the performance in one direction.
  • In quality-of-service (QoS) enabled networks, provisioning in one direction may be radically different from provisioning in the reverse direction, so the QoS guarantees differ.

on the other hand, round-trip delay measurement has two specific advantages, and they're iconic 🙈

  • Ease of deployment: unlike one-way measurement, it's often possible to do some form of round-trip delay measurement without installing measurement-specific software at the intended destination. A variety of approaches are well-known, including use of ICMP Echo or of TCP-based methodologies (similar to those outlined in "IPPM Metrics for Measuring Connectivity" [4]). However, some approaches may introduce greater uncertainty in the time for the destination to produce a response (see Section 2.7.3).
  • Ease of interpretation: in some circumstances, the round-trip time is actually the quantity of interest. Deducing the round-trip time from matching one-way measurements and an assumption about the destination processing time is less direct and potentially less accurate.

1.2. General Issues Regarding Time ⏰

whenever a time (i.e., a moment in history) is mentioned here, it's understood to be measured in seconds (and fractions) relative to UTC 🌍

as the Framework document explains more fully, there are four distinct but related notions of clock uncertainty 🧚

synchronization* measures the extent to which two clocks agree on what time it is. For example, the clock on one host might be 5.4 msec ahead of the clock on a second host.

accuracy* measures the extent to which a given clock agrees with UTC. For example, the clock on a host might be 27.1 msec behind UTC.

resolution* measures the precision of a given clock. For example, the clock on an old Unix host might tick only once every 10 msec, and thus have a resolution of only 10 msec.

skew* measures the change of accuracy, or of synchronization, with time. For example, the clock on a given host might gain 1.3 msec per hour and thus be 27.1 msec behind UTC at one time and only 25.8 msec an hour later. In this case, we say that the clock of the given host has a skew of 1.3 msec per hour relative to UTC, which threatens accuracy. We might also speak of the skew of one clock relative to another clock, which threatens synchronization.

2. A Singleton Definition for Round-trip Delay 🦋

2.1. Metric Name:

the name is Type-P-Round-trip-Delay 💖

2.2. Metric Parameters:

  • Src, the IP address of a host
  • Dst, the IP address of a host
  • T, a time

2.3. Metric Units:

the value of a Type-P-Round-trip-Delay is either a real number, or an undefined (informally, infinite) number of seconds 🔥

2.4. Definition:

okay this is the important part, so pay attention bestie 👑

For a real number dT, >>the Type-P-Round-trip-Delay from Src to Dst at T is dT<< means that Src sent the first bit of a Type-P packet to Dst at wire-time* T, that Dst received that packet, then immediately sent a Type-P packet back to Src, and that Src received the last bit of that packet at wire-time T+dT.

The Type-P-Round-trip-Delay from Src to Dst at T is undefined (informally, infinite)<< means that Src sent the first bit of a Type-P packet to Dst at wire-time T and that (either Dst did not receive the packet, Dst did not send a Type-P packet in response, or) Src did not receive that response packet.
The Type-P-Round-trip-Delay between Src and Dst at T<< means either the **Type-P-Round-trip-Delay from Src to Dst at T or the Type-P-Round-trip-Delay from Dst to Src at T. When this notion is used, it's understood to be specifically ambiguous which host acts as Src and which as Dst. {Comment: This ambiguity will usually be a small price to pay for being able to have one measurement, launched from either Src or Dst, rather than having two measurements.}

suggestions for what to report along with metric values appear in Section 3.8, after a discussion of the metric, methodologies for measuring it, and error analysis 🧚

2.5. Discussion:

Type-P-Round-trip-Delay is a relatively simple analytic metric, and we believe it will afford effective methods of measurement 💅

the following issues are likely to come up in practice 🙈

  • The timestamp values (T) for the time at which delays are measured should be fairly accurate, so you can draw meaningful conclusions about the state of the network at a given T. Therefore, Src should have an accurate knowledge of time-of-day. NTP [3] is one way to get time accuracy to within several milliseconds. Depending on the NTP server, higher accuracy may be achieved, for example when NTP servers make use of GPS systems as a time source. Note that NTP will adjust the instrument's clock. If an adjustment is made between the time the initial timestamp is taken and the time the final timestamp is taken, the adjustment will affect the uncertainty in the measured delay. This uncertainty must be accounted for in the instrument's calibration.
  • A given methodology will have to include a way to determine whether a delay value is infinite or whether it's merely very large (and the packet is yet to arrive at Dst). As noted by Mahdavi and Paxson [4], simple upper bounds (such as the 255 seconds theoretical upper bound on the lifetimes of IP packets [5]) could be used, but good engineering, including an understanding of packet lifetimes, will be needed in practice. {Comment: Note that, for many applications of these metrics, the harm in treating a large delay as infinite might be zero or very small. A TCP data packet, for example, that arrives only after several multiples of the RTT may as well have been lost.}
  • If the packet is duplicated so that multiple non-corrupt instances of the response arrive back at the source, then the packet is counted as received, and the first instance to arrive back at the source determines the packet's round-trip delay.
  • If the packet is fragmented and, for whatever reason, reassembly does not occur, then the packet will be deemed lost.

2.6. Methodologies: ✨

okay so, as with every other Type-P-* metric, the actual methodology depends on the Type-P 🎀 (think protocol number, UDP/TCP port number, size, precedence). it's literally the whole vibe of the test 💅

generally, for a given Type-P, the flow goes like this:

  • at the Src host, pick your Src and Dst IP addresses and build a test packet of Type-P with those addresses 🦋. any 'padding' part that's only there to make the packet a certain size should be filled with randomized bits 🎲, so the measured delay isn't lower than it'd be otherwise because of compression tricks along the path 🙈. the test packet MUST have some identifying info so Src can match up the response when it comes back 🫶. one way to do that is to stick the timestamp taken just before sending the test packet inside the packet itself 🌸.
  • at the Dst host, get it set up to receive and respond to the test packet 🧚. at the Src host, get it set up to receive the matching response packet 💜.
  • at the Src host, take the initial timestamp and then send the prepared Type-P packet toward Dst 🚀. the timestamp could live inside the packet, or be kept separately, as long as the packet has a suitable identifier so the received timestamp can be compared with the send timestamp 🔥.
  • if the packet arrives at Dst, send a matching response packet from Dst back to Src as soon as possible 🏃‍♀️💨.
  • if the response packet arrives within a reasonable period of time, take the final timestamp as soon as possible after the packet is received 💖. subtract the two timestamps and you get an estimate of round-trip delay ✨. if the delay between the initial timestamp and the actual sending of the packet is known, the estimate could be adjusted by subtracting that amount, but the uncertainty in that value MUST be taken into account in the error analysis 🧐. same deal if the delay between the actual receipt of the response packet and the final timestamp is known: the estimate could be adjusted by subtracting it, and the uncertainty in that value MUST be taken into account in the error analysis 🙌. see the next section, "Errors and Uncertainties", for the deeper dive 🌈.
  • if the packet fails to arrive within a reasonable period of time, the round-trip delay is taken to be undefined (informally, infinite) 💀. note that the threshold of 'reasonable' is a parameter of the methodology 🍓.

issues like the packet format and how Dst knows when to expect the test packet are outside the scope of this document 📦.

{Comment: note that you cannot in general add two Type-P-One-way-Delay values (see [2]) to form a Type-P-Round-trip-Delay value. to form a Type-P-Round-trip-Delay value, the return packet MUST be triggered by the reception of a packet from Src.} 🙃

{Comment: "ping" would qualify as a round-trip measure under this definition, with a Type-P of ICMP echo request/reply with 60-byte packets. however, the uncertainties associated with a typical ping program MUST be analyzed as in the next section, including the type of reflecting point (a router may not handle an ICMP request in the fast path) and effects of load on the reflecting point.} 😭

2.7. Errors and Uncertainties: 🔍

the description of any specific measurement method SHOULD include an accounting and analysis of the different sources of error or uncertainty 📝. the Framework document gives general guidance on this, but here's the delay-specific stuff we need to flag 💅:

  • errors or uncertainties from the uncertainty in the clock of the Src host ⏰.
  • errors or uncertainties from the difference between 'wire time' and 'host time' 🖥️.
  • errors or uncertainties from the time the Dst needs to receive the packet from Src and send the matching response 🏓.

also, the loss threshold can affect the results 🚨. each of these gets a closer look below, plus a section ("Calibration") on accounting for them 🎀.

the uncertainty in a round-trip delay measurement is partly related to the uncertainty in the clock of the Src host 🕰️. below, the clock used to measure when the packet was sent from Src is the source clock 🫧. the observed time when the packet was sent is Tinitial, and the observed time when it was received by the source is Tfinal 🌷. tying back to the notions of synchronization, accuracy, resolution, and skew from the Introduction, here's the tea ☕:

  • for one-way delay there's an issue with synchronizing the source clock and the destination clock 🙅‍♀️. for round-trip delay it's an (easier) issue of self-synchronization, like the source clock has to agree with itself between the moment the test packet goes out and the moment the response comes in 🪞. theoretically, a really severe case of skew could threaten this 😬. in practice, the bigger threat is anything that causes a discontinuity in the source clock between taking the initial and final timestamp 🫠. this might happen, for example, with certain implementations of NTP 🙈.
  • accuracy of a clock only matters for figuring out when a given delay was measured 📅. accuracy has no real bearing on how accurate the delay measurement itself is 🤷‍♀️.
  • resolution of a clock adds uncertainty to any time measured with it 📏. so if the source clock has a resolution of 10 msec, that adds 10 msec of uncertainty to any time value measured with it 💥. we'll denote the resolution of the source clock as Rsource 🏷️.

putting it all together, the naive computation Tfinal-Tinitial will be off by 2*Rsource 🫶.

as we defined round-trip delay, we'd really like to measure the time between when the test packet leaves the network interface of Src and when the matching response packet (completely) arrives at the network interface of Src 🛫🛬. we call these "wire times" 🧵. but if software on Src does the timing, that software can only directly measure the time between when Src grabs a timestamp just before sending the test packet and when it grabs one just after receiving the response 🕹️. we call these two points "host times" 🏠.

another contributor to this problem is the time spent at Dst between receiving the test packet and sending the response packet 🕐. ideally, this time is zero 🙏. it gets explored further in the next section 🔎.

to the extent the difference between wire time and host time is accurately known, that knowledge can be used to correct host time measurements, and the corrected value more accurately estimates the desired (wire time) metric 🎯.

to the extent the difference between wire time and host time is uncertain, that uncertainty MUST be accounted for in the analysis of a given measurement method 🧠. we denote by Hinitial an upper bound on the uncertainty in the difference between wire time and host time on the Src host when sending the test packet 📤, and similarly define Hfinal for the difference on the Src host when receiving the response packet 📥. these problems together introduce a total uncertainty of Hinitial + Hfinal 🧮. this estimate of total wire-vs-host uncertainty SHOULD be included in the error/uncertainty analysis of any measurement implementation 💌.

any time the destination host spends receiving and recognizing the packet from Src, then producing and sending the matching response, adds extra error and uncertainty to the round-trip delay measurement 😤. the error equals the difference between the wire time the first bit of the packet is received by Dst and the wire time the first bit of the response is sent by Dst 🧩. to the extent that this difference is accurately known, that knowledge can be used to correct the desired metric 🩹. to the extent it's uncertain, that uncertainty MUST be accounted for in the error analysis of a measurement implementation 🥺. we denote this uncertainty by Hrefl 🏷️. this estimate of uncertainty SHOULD be included in the error/uncertainty analysis of any measurement implementation 🫶.

we're getting so close to the good stuff, bestie 🌈✨💖

2.7.4. Calibration 🧁

Okay so, generally the measured values can be broken down like this, bestie 🫶:

measured value = true value + systematic error + random error

If the systematic error (the constant bias in measured values) can be determined, it can be compensated for in the reported results. ✨

reported value = measured value - systematic error

therefore

reported value = true value + random error

The goal of calibration is to figure out the systematic and random error the instruments themselves generate, in as much detail as possible. 💅 At a minimum, a bound (e) should be found so that the reported value is in the range (true value - e) to (true value + e) at least 95 percent of the time. We call e the calibration error for the measurements. It shows how repeatable the values from the measurement instrument are, like how closely an actual delay of 30 ms gets reported as 30 ms. {Comment: 95 percent was chosen because (1) some confidence level is desirable to be able to remove outliers, which will be found in measuring any physical property; and (2) a particular confidence level should be specified so that the results of independent implementations can be compared.} 🌸

From the discussion in the previous three sections, the error in measurements could be bounded by working out all the individual uncertainties and adding them together, like this:

2*Rsource + Hinitial + Hfinal + Hrefl.

Still, reasonable bounds on both the clock-related uncertainty captured by the first term and the host-related uncertainty captured by the last three terms should be possible, through careful design techniques and by calibrating the instruments on a known, isolated network in a lab. 🔬

The host-related uncertainties, Hinitial + Hfinal + Hrefl, could be bounded by connecting two instruments back-to-back with a high-speed serial link or an isolated LAN segment. 🔌 In this case, repeated measurements are measuring the same round-trip delay. 🔁

If the test packets are small, a network connection like that has a minimal delay that may be approximated by zero. So the measured delay only contains systematic and random error from the instrumentation. The "average value" of repeated measurements is the systematic error, and the variation is the random error. 📊

One way to compute the systematic error, and the random error to a 95% confidence, is to repeat the experiment many times, at least hundreds of tests. 🔁 The systematic error would then be the median. The random error could then be found by removing the systematic error from the measured values. The 95% confidence interval would be the range from the 2.5th percentile to the 97.5th percentile of these deviations from the true value. The calibration error e could then be taken to be the largest absolute value of these two numbers, plus the clock-related uncertainty. {Comment: as described, this bound is relatively loose since the uncertainties are added, and the absolute value of the largest deviation is used. As long as the resulting value is not a significant fraction of the measured values, it is a reasonable bound. If the resulting value is a significant fraction of the measured values, then more exact methods will be needed to compute the calibration error.} 🙈

Note that random error is a function of measurement load. 🧠 For example, if many paths will be measured by one instrument, this might increase interrupts, process scheduling, and disk I/O (for example, recording the measurements). All of that may increase the random error in measured singletons. So, in addition to minimal load measurements to find the systematic error, calibration measurements should be performed with the same measurement load the instruments will see in the field. 🎀

We want to say it again, because it matters: this statistical treatment is about calibrating the instrument. It's how we "calibrate the meter stick" and say how well the meter stick reflects reality. 📏✨

In addition to calibrating the instruments for finite delay, two checks should be made to make sure packets reported as losses were really lost. 🕵️‍♀️ First, the threshold for loss should be verified. In particular, make sure the "reasonable" threshold is actually reasonable: it should be very unlikely that a packet arrives after the threshold value, so the number of packets lost over an interval isn't sensitive to the error bound on measurements. Second, consider the possibility that a packet arrives at the network interface but is lost due to congestion on that interface, or to other resource exhaustion (e.g. buffers) in the instrument. 💀

2.8. Reporting the Metric: 📝

The calibration and the context the metric is measured in MUST be carefully considered, and SHOULD always be reported along with metric results. 🫶 We now present four items to consider: the Type-P of test packets, the threshold of infinite delay (if any), error calibration, and the path traversed by the test packets. This list is not exhaustive. Any additional information that could be useful in interpreting applications of the metrics should also be reported. 🦋

2.8.1. Type-P 🎀

As noted in the Framework document [1], the value of the metric may depend on the type of IP packets used to make the measurement, or "type-P". 💖 The value of Type-P-Round-trip-Delay could change if the protocol (UDP or TCP), port number, size, or arrangement for special treatment (e.g., IP precedence or RSVP) changes. The exact Type-P used to make the measurements MUST be accurately reported. 📣

2.8.2. Loss threshold 🚫

In addition, the threshold (or methodology to distinguish) between a large finite delay and loss MUST be reported. ⏱️

2.8.3. Calibration Results 📈

  • If the systematic error can be determined, it SHOULD be removed from the measured values. ✨
  • You SHOULD also report the calibration error, e, such that the true value is the reported value plus or minus e, with 95% confidence (see the last section.) 🌈
  • If possible, the conditions under which a test packet with finite delay is reported as lost due to resource exhaustion on the measurement instrument SHOULD be reported. 🧃

2.8.4. Path 🛣️

Finally, the path traversed by the packet SHOULD be reported, if possible. 🗺️ In general it's impractical to know the precise path a given packet takes through the network. The precise path may be known for certain Type-P on short or stable paths. For example, if Type-P includes the record route (or loose-source route) option in the IP header, the path is short enough, all routers* on the path support record (or loose-source) route, and the Dst host copies the path from Src to Dst into the corresponding reply packet, then the path will be precisely recorded. 🧩

Honestly though, this is impractical. The route has to be short enough, many routers don't support (or aren't configured for) record route, and using this feature would often artificially worsen the performance observed by pulling the packet out of common-case processing. 😤 Partial information is still valuable context, though! For example, if a host can choose between two links* (and hence two separate routes from Src to Dst), then the initial link used is valuable context. {Comment: For example, with Merit's NetNow setup, a Src on one NAP can reach a Dst on another NAP by either of several different backbone networks.} 👑

3. A Definition for Samples of Round-trip Delay ✨

Given the singleton metric Type-P-Round-trip-Delay, we now define one particular sample of those singletons. 🧁 The idea of the sample is to pick a particular binding of the parameters Src, Dst, and Type-P, then define a sample of values of parameter T. To define the values of T, you select a beginning time T0, a final time Tf, and an average rate lambda, then define a pseudo-random Poisson process of rate lambda whose values fall between T0 and Tf. The time interval between successive values of T will then average 1/lambda. 🦋

{Comment: Note that Poisson sampling is only one way of defining a sample. Poisson has the advantage of limiting bias, but other methods of sampling might be appropriate for different situations. We encourage others who find such appropriate cases to use this general framework and submit their sampling method for standardization.} 🌸

3.1. Metric Name: 🏷️

Type-P-Round-trip-Delay-Poisson-Stream

3.2. Metric Parameters: 🎛️

  • Src, the IP address of a host 💻
  • Dst, the IP address of a host 🖥️
  • T0, a time ⏰
  • Tf, a time ⏳
  • lambda, a rate in reciprocal seconds 🔢

3.3. Metric Units: 📏

A sequence of pairs. The elements of each pair are:

  • T, a time, and 🕒
  • dT, either a real number or an undefined number of seconds. 💜

The values of T in the sequence are monotonic increasing. Note that T would be a valid parameter to Type-P-Round-trip-Delay, and that dT would be a valid value of Type-P-Round-trip-Delay. 🧠

3.4. Definition: 📖

Given T0, Tf, and lambda, we compute a pseudo-random Poisson process beginning at or before T0, with average arrival rate lambda, and ending at or after Tf. Those time values greater than or equal to T0 and less than or equal to Tf are then selected. At each of the times in this process, we obtain the value of Type-P-Round-trip-Delay at that time. The value of the sample is the sequence made up of the resulting <time, delay> pairs. If there are no such pairs, the sequence is of length zero and the sample is said to be empty. 🫧

3.5. Discussion: 💬

The reader should be familiar with the in-depth discussion of Poisson sampling in the Framework document [1], which includes methods to compute and verify the pseudo-random Poisson process. 📚

We specifically do not constrain the value of lambda, except to note the extremes. If the rate is too large, then the measurement traffic will perturb the network, and itself cause congestion. 🚦 If the rate is too small, then you might not capture interesting network behavior. {Comment: We expect to document our experiences with, and suggestions for, lambda elsewhere, culminating in a "best current practices" document.} 🌟

Since a pseudo-random number sequence is employed, the sequence of times, and hence the value of the sample, is not fully specified. Pseudo-random number generators of good quality will be needed to achieve the desired qualities. 🎲

The sample is defined in terms of a Poisson process both to avoid the effects of self-synchronization and to capture a sample that is statistically as unbiased as possible. {Comment: there is, of course, no claim that real Internet traffic arrives according to a Poisson arrival process.} The Poisson process is used to schedule the delay measurements. The test packets will generally not arrive at Dst according to a Poisson distribution, nor will response packets arrive at Src according to a Poisson distribution, since they are influenced by the network. 🌐

All the singleton Type-P-Round-trip-Delay metrics in the sequence will have the same values of Src, Dst, and Type-P. 👯‍♀️

Note also that, given one sample that runs from T0 to Tf, and given new time values T0' and Tf' such that T0 <= T0' <= Tf' <= Tf, the subsequence of the given sample whose time values fall between T0' and Tf' is also a valid Type-P-Round-trip-Delay-Poisson-Stream sample. 🎀

3.6. Methodologies 🎀

okay so the methodologies just follow straight from two things:

  • the way we pick specific times, using the specified Poisson arrival process, and
  • the methodologies discussion we already did for the singleton Type-P-Round-trip-Delay metric 🧚

bestie, we have to be super careful about out-of-order arrival of test or response packets 😭. The Src could send one test packet at TS[i], then send a second test packet (later) at TS[i+1], and then receive the second response packet at TR[i+1]. Then the first response packet shows up (later) at TR[i]. Plot twist, the packets literally swapped places on the way back 💀

3.7. Errors and Uncertainties 🦋

so on top of the errors and uncertainties we already talked about for measuring the singleton values in the sample, we also have to check how accurate the Poisson process is with respect to the wire-times of sending the test packets 👀. Things that can mess this up include problems with the pseudo-random number techniques used to make the Poisson arrival process, or jitter in the value of Hinitial (the uncertainty we mentioned in the singleton delay metric). The Framework document shows how to use the Anderson-Darling test to verify the accuracy of a Poisson process over small time frames 🌸

{Comment: The goal is to ensure that test packets are sent "close enough" to a Poisson schedule, and avoid periodic behavior.} ✨ basically we want the packets to feel random but not too random, and definitely not on a boring repeating schedule 💅

3.8. Reporting the Metric 💖

you MUST report the calibration and context for the underlying singletons along with the stream 🫶 (See "Reporting the metric" for Type-P-Round-trip-Delay.) no skipping this part, it's giving important 🔥

4. Some Statistics Definitions for Round-trip Delay 📊

okay so now that we have the sample metric Type-P-Round-trip-Delay-Poisson-Stream, we can offer some stats about it ✨. These stats are mostly here to be illustrative of what could be done, so think of them as inspo, not rules 🧚

4.1. Type-P-Round-trip-Delay-Percentile 🌈

given a Type-P-Round-trip-Delay-Poisson-Stream and a percent X between 0% and 100%, we get the Xth percentile of all the dT values in the Stream. when we compute this, undefined values get treated as infinitely large 🙈. Which means the percentile could be undefined (informally, infinite), so no cap, it can be giving infinity 💀. Also, the Type-P-Round-trip-Delay-Percentile is undefined if the sample is empty 🫠

example time! suppose we take a sample and the results are:

Stream1 = <
<T1, 100 msec>
<T2, 110 msec>
<T3, undefined>
<T4, 90 msec>
<T5, 500 msec>
>

then the 50th percentile would be 110 msec 💖, since 90 msec and 100 msec are smaller, and 110 msec and 'undefined' are larger.

also, if there's a significant chance that a packet with finite delay gets reported as lost, then a high percentile (90th or 95th) might get reported as infinite instead of finite 😭 which is lowkey wild.

4.2. Type-P-Round-trip-Delay-Median 🍓

given a Type-P-Round-trip-Delay-Poisson-Stream, the median is the median of all the dT values in the Stream 🧁. Undefined values are treated as infinitely large here too. Same as with the percentile, Type-P-Round-trip-Delay-Median is undefined if the sample is empty 🙃

as the Framework document says, the median only differs from the 50th percentile when the sample has an even number of values. In that case we take the mean of the two central values ✨

example time again:

Stream2 = <
<T1, 100 msec>
<T2, 110 msec>
<T3, undefined>
<T4, 90 msec>
>

so the median would be 105 msec 🎀, the mean of 100 msec and 110 msec, the two central values. iconic math moment 🧮

4.3. Type-P-Round-trip-Delay-Minimum 🦋

given a Type-P-Round-trip-Delay-Poisson-Stream, the minimum is the smallest of all the dT values in the Stream 🌸. Undefined values are treated as infinitely large when we compute this. Which means the minimum could be undefined (informally, infinite) if all the dT values are undefined 😭. And the Type-P-Round-trip-Delay-Minimum is undefined if the sample is empty 💅

in the example above, the minimum would be 90 msec 💜

4.4. Type-P-Round-trip-Delay-Inverse-Percentile 👑

given a Type-P-Round-trip-Delay-Poisson-Stream and a time duration threshold, the Inverse-Percentile is the fraction of all the dT values in the Stream that are less than or equal to the threshold 🧚. The result could be as low as 0% (if all the dT values exceed the threshold) or as high as 100% 🌈. Type-P-Round-trip-Delay-Inverse-Percentile is undefined if the sample is empty 🫠

in the example above, the Inverse-Percentile of 103 msec would be 50% ✨

5. Security Considerations 🔐

okay bestie, this section is SERIOUS, so please read it properly 🙏.

Conducting Internet measurements raises both security and privacy concerns 😬. This memo doesn't specify an implementation of the metrics, so it doesn't directly affect the security of the Internet or of applications running on it. Still, implementations of these metrics must be mindful of security and privacy concerns 🫶

there are two types of security concerns: potential harm caused by the measurements, and potential harm to the measurements 💀. The measurements could cause harm because they're active, and they inject packets into the network. The measurement parameters MUST be carefully selected so that the measurements inject trivial amounts of additional traffic into the networks they measure 🚨. If they inject "too much" traffic, they can skew the results of the measurement, and in extreme cases cause congestion and denial of service. not it 😩

the measurements themselves could be harmed by routers giving measurement traffic a different priority than "normal" traffic, or by an attacker injecting artificial measurement traffic 👀. If routers can recognize measurement traffic and treat it separately, the measurements won't reflect actual user traffic. And if an attacker injects artificial traffic that gets accepted as legitimate, the loss rate will be artificially lowered 📉. So the measurement methodologies SHOULD include appropriate techniques to reduce the probability that measurement traffic can be distinguished from "normal" traffic 🕵️‍♀️. Authentication techniques, such as digital signatures, may be used where appropriate to guard against injected traffic attacks 🔐

the privacy concerns of network measurement are limited by the active measurements described in this memo 🌷. Unlike passive measurements, there can be no release of existing user data. so that's a W for privacy 💖

6. Acknowledgements 🫶

special thanks are due to Vern Paxson and to Will Leland for several useful suggestions 🧚💜

7. References 📚

[1] Paxson, D., Almes, G., Mahdavi, J. and M. Mathis, "Framework for IP Performance Metrics", RFC 2330, May 1998.

[2] Almes, G., Kalidindi,S. and M. Zekauskas, "A One-way Delay Metric for IPPM", RFC 2679, September 1999.

[3] Mills, D., "Network Time Protocol (v3)", RFC 1305, April 1992.

[4] Mahdavi, J. and V. Paxson, "IPPM Metrics for Measuring Connectivity", RFC 2678, September 1999.

[5] Postel, J., "Internet Protocol", STD 5, RFC 791, September 1981.

[6] Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, March 1997.

8. Authors' Addresses 📮

Guy Almes
Advanced Network & Services, Inc.
200 Business Park Drive
Armonk, NY  10504
USA

Phone: +1 914 765 1120
EMail: almes@advanced.org


Sunil Kalidindi
Advanced Network & Services, Inc.
200 Business Park Drive
Armonk, NY  10504
USA

Phone: +1 914 765 1128
EMail: kalidindi@advanced.org


Matthew J. Zekauskas
Advanced Network & Services, Inc.
200 Business Park Drive
Armonk, NY 10504
USA

Phone: +1 914 765 1112
EMail: matt@advanced.org

Copyright (C) The Internet Society (1999). All Rights Reserved. 💜

This document and translations of it may be copied and furnished to others, and derivative works that comment on or otherwise explain it or assist in its implementation may be prepared, copied, published and distributed, in whole or in part, without restriction of any kind, provided that the above copyright notice and this paragraph are included on all such copies and derivative works. However, this document itself may not be modified in any way, such as by removing the copyright notice or references to the Internet Society or other Internet organizations, except as needed for the purpose of developing Internet standards in which case the procedures for copyrights defined in the Internet Standards process must be followed, or as required to translate it into languages other than English.

The limited permissions granted above are perpetual and will not be revoked by the Internet Society or its successors or assigns.

This document and the information contained herein is provided on an "AS IS" basis and THE INTERNET SOCIETY AND THE INTERNET ENGINEERING TASK FORCE DISCLAIMS ALL WARRANTIES, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE INFORMATION HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.

Acknowledgement 🌈

Funding for the RFC Editor function is currently provided by the Internet Society. 🎀

📦 where this page comes from