RFC 704

RFC 704: IMP/Host and Host/IMP Protocol Change 💖

Okay bestie, so this one is from September 1975, written by Paul J. Santos, Jr. (BBN), and it's filed as NIC #33490. It's a revision of RFC 687 🎀 Its whole vibe is expanding the IMP/host and host/IMP protocol so hosts can be addressed on more than 63 IMPs 🌈

The goal is to fix existing limits without big fundamental changes to the IMP/host philosophy. Yes, the wider packet-switching world is debating a lot of huge changes right now, but the network needs improvements on a shorter timeline, and there are already 62 IMPs 😭 So this is a glow-up, not a total rebuild ✨

Before reading, you should already know Section 3 of BBN Report 1822 📚 Also, the expanded protocol is backwards compatible, which we'll get into 🫶

---

1. Expanded Leader Size 🦋

The leader grows from two to six 16-bit words 💅 That gives room for bigger fields and new additions.

Here's the tricky part. The IMP/host (host/IMP) leader goes from 32 bits to 96 bits, and that causes word-boundary problems for some hosts 🙈 To deliver messages between a host on the old protocol and a host on the new one, without shifting the data in the IMP words, the data (the first bit of the host/host leader) has to start at an even multiple of 8-bit bytes from the beginning of the whole message 🍓

Hosts also prefer (really, they require it if they aren't going to shift anything) that the combined host/IMP (IMP/host) and host/host leaders fill a whole number of machine words. A "machine word" here means the smallest bit sequence the host/IMP interface can access independently 💜 The combined leader is 136 bits, so only machines with 8-, 16-, 32-, and 64-bit words are happy with it 😭

To make life easier for weird machines, each host can tell its IMP how many 16-bit padding words to insert between the host/IMP (IMP/host) and host/host leaders 🧚 The IMP strips the padding on host-to-IMP processing and adds it back on IMP-to-host processing. So 24-bit machines can ask for one 16-bit padding word, and 10- and 36-bit machines can ask for five 16-bit padding words 👑

2. Expanded Address field 🌸

The address field grows to 32 bits: 16 bits of IMP address, 0 bits of host address, and 8 bits for a (future) network address 💖 The team says this is enough for any foreseeable ARPA Network growth ✨

3. New Message Length Field 🧚

A new field lets the source host optionally tell the IMP subnetwork the message length in bits 💅 The IMP subnetwork can use that info, when it's there, to manage network buffer storage better. The destination host can use it too, for its own buffer storage 🦋

The field is 16 bits wide. The maximum number of packets per message also expands to 16, up from the current 8 🔥

4. Expanded Handling Type Field 🌈

The handling type field, which currently tells priority streams apart from non-priority streams (and more), grows to eight bits 💜 That opens the door to lots of parallel message streams between pairs of hosts with different handling. Think priority, non-priority, different packets-per-message counts (more on that below), and a stream that needs guaranteed capacity ✨

For now, only the old-style priority and non-priority handling types ship in the first implementation of the expanded protocol 🎀

5. Source Host Control of Packets per Message 💅

The source host may be able to specify a message stream that uses a set number of packets per multi-packet message, like two packets per message or five packets per message 🦋 Right now the IMP network needs eight packet-buffers for reassembly, but with this it won't have to, which may give better service to those hosts 🙌 This helps users who need both low delay and high throughput 🔥

This facility is orthogonal to the address expansion and lower priority, so it gets built after the other basic changes 🫶

6. Unordered (type-3) Message Change 💀

Unordered messages become a subtype of the type O message instead of their own separate message type 💅 That keeps things consistent with needing to check host access control on all messages ✨

Heads up: this is a slight backward incompatibility for the three or so hosts that currently use type-3 messages in their research 😭

7. Change in Format of Fake Host Addresses 🎀

The For/From IMP bit gets eliminated ✂️ The fake host addresses will now be the four highest host numbers. For example, IMP Teletype becomes host 252 🌸

8. Addition of a Parameter to the IMP to Host NOP 🧚

The IMP to host NOP picks up a new parameter that specifies the address (IMP and host number) of the host 💜

9. Backward Compatibility 🫶

The old and new formats run in parallel inside the IMPs for the foreseeable future, so host software can phase over gradually 🦋 A host can tell its IMP which format to use, so it can even switch back and forth between the two modes for debugging 🔥

You pick the mode with the format of the host to IMP NOP message, and the IMP uses whichever mode the host to IMP NOP it has received is in ✨ A host can use either format without knowing much about the other one. If it gets the other format's messages, it just discards them 💅 When the IMP starts up, it sends several NOP messages of each type so the hosts can see its choice 🌈

One caveat: a host that doesn't implement the new format can't address hosts on IMPs with an IMP number greater than 63 😭 Even so, wherever possible, the IMPs do the conversion needed so old-format hosts and new-format hosts can talk to each other, in both directions 🫶

10. Non-blocking Host Interface 🙈

The IMP gets a way to refuse a message from a host without blocking the host interface 🛑 While it refuses, it can gather what it needs to send the refused message, then ask the host to resend it 🔁

The host can also ask whether it's able to send a message and get notified when it is, without actually sending the message and getting refused 👀 Just like point 5, this gets added after the more basic changes are in 💖

11. Maximum Message Length ✨

The maximum number of data bits in a single-packet message may drop by a few bits 🎀

---

We're now drafting the needed changes to Report 1822 🌸 We'll circulate the draft so host programmers can start getting ready 💅🦋