RFC 557

REVELATIONS IN NETWORK HOST MEASUREMENTS 💖✨

• RFC: 557
• Title: Revelations in Network Host Measurements
• Author: B. Wessler, Telenet Communications 🎀
• Series: Network Working Group, NIC 18457
• Date: 30 August 1973 📅
• Category, obsoletes, updates: none listed in the original 🌸
• References: RFC 392, RFC 415 🦋

Okay bestie, so RFC 392 was supposed to point out a problem the Utah folks hit while using the ARPANET 🫶. The big idea was "here's a place to make the Network better," NOT "look how terrible the Network is" 💀. Sadly, people read it as the second one, which is so relatable 😭. The other goal was to get people measuring stuff on other computers and operating systems too, not just TENEX. Barely anyone did, except Hal Murray with RFC 415, who measured TENEX 💅.

Since RFC 392 came out, BBN has been on a glow-up journey with Host-Net performance on TENEX ✨. They shared new numbers in their April 1973 Quarterly Progress Report No. 10, and I think the RFC community needs to see them 👑.

Here's the tea 🍓. Don Allen at BBN took Greg Hicks' RJS program and did three things: 1) updated it for recent TENEX changes, 2) improved the code near the input/output JSYs, and 3) used much faster network monitor code. The result was about a 400% improvement 🔥. Cost went from 75-85 seconds of CPU time per megabit (~$10) down to 19 seconds per megabit (~$2.50) 🤑. Of those 19 seconds, 13 were in the RJS program and 6 were in TENEX network output 🧚.

Those six seconds line up really well with BBN FTP needs, which took 8.2 cpu seconds per megabit (~$1.08) for 8 bit byte transfers 💜. Switching to 32 or 36 bit bytes improves that by a factor of 4, which brings it to $.33 per megabit 🌈.

As for the other 13 seconds in RJS, nobody tried to improve them or even figure out where the time was going 🙈. Honestly, that's fair, because RJS is about to be replaced by the RJE protocol, which uses FTP as its transfer mechanism 🦋.

So the vibe check is this: the original RFC 392 and the recent BBN results show that the Network, Host cost included, is intrinsically effective 🫶. If you're not careful with monitor and user code, though, the system can look pretty unattractive 😬. I really hope everyone goes out and measures how good (or bad) they're doing vis a vis network transfers 📊. If the results are too embarrassing to publish as an RFC, send them to me personally 🙈. It'd be so nice to hear from all the systems 💅.

All the data is courtesy of Don Allen and Jerry Burchfiel at BNN 🌸.

[ This RFC was put into machine readable form for entry ] [ into the online RFC archives by Lorrie Shiota ] 🎀