Cross Country Network Bandwidth
A back-of-envelope count showing one cross-country keystroke costs four packets each way.
Deutsch, L.P., "Cross Country Network Bandwidth", RFC 567, DOI 10.17487/RFC0567, September 1973, <https://www.rfc-editor.org/info/rfc567>.
@misc{rfc567,
series = {Request for Comments},
number = 567,
howpublished = {RFC 567},
publisher = {RFC Editor},
doi = {10.17487/RFC0567},
url = {https://www.rfc-editor.org/info/rfc567},
author = {L.P. Deutsch},
title = {{Cross Country Network Bandwidth}},
pagetotal = 1,
year = 1973,
month = sep,
} RFC text
Network Working Group L. Peter Deutsch (PARC-MAXC)
Request for Comments: 567 September 6, 1973
NIC #18970
CROSS-COUNTRY NETWORK BANDWIDTH
The following computation of cross-country network bandwidth was
contributed by Butler Lampson of PARC.
Consider what happens when a TIP user on the West Coast, connected to a
full-duplex Host on the East Coast, strikes a key on his terminal.
The TIP sends a one-character message (1 packet).
The destination IMP sends a RFNM (1 packet).
The destination Host sends an ALLocate - this seems to be the strategy
used by TENEX Hosts, at least (1 packet).
Thc TIP sends a RFNM for the ALLocate (1 packet).
The same sequence repeats itself, with roles interchanged, for the echo
character (4 packets).
This constitutes 4 packets or 4OOO bits in each direction. The current
cross-country transmission capability of the ARPANET is 3 5OKb phone
lines; ergo, it can only support 3*50000/4000=37.5 such characters per
second!
It may be that RFNMs are transmitted between IMPs more efficiently; at
best this can only double the network capacity.
This computation may help explain why cross-country TIP users (e.g. the
substantial West Coast community of BBN-TENEX users) experience such
bad echo response, at least in bursts: the network itself may be
experiencing momentary peak loads.
If this argument is correct, the proposed remote echoing facilities of
the new TELNET protocol could have a major effect on network operation.
AI ELI5
On this page
§1One Keystroke, Many Packets
L. Peter Deutsch passes on a computation contributed by Butler Lampson of PARC. A TIP user on the West Coast, connected to a full-duplex Host on the East Coast, strikes a key.
The TIP sends a one-character message (1 packet). The destination IMP sends a RFNM (1 packet). The destination Host sends an ALLocate, which the author says seems to be the strategy used by TENEX Hosts (1 packet). The TIP sends a RFNM for the ALLocate (1 packet).
One key press makes a small message, a receipt, a note saying the computer is ready for more, and a receipt for that note.
§2The Arithmetic
The same sequence repeats with roles swapped for the echo character, so the exchange is 4 packets, or 4000 bits, in each direction. The cross-country capability of the ARPANET was 3 50Kb phone lines, so it can support only 3*50000/4000 = 37.5 such characters per second.
If RFNMs are transmitted between IMPs more efficiently, the capacity could at best double.
Add up all the little messages and the cross-country road only fits about 37 key presses each second.
§3Why It Matters
The author suggests this may explain why cross-country TIP users, such as the large West Coast community of BBN-TENEX users, see bad echo response in bursts: the network itself may be experiencing momentary peak loads.
If the argument is correct, the proposed remote echoing facilities of the new TELNET protocol could have a major effect on network operation.
This is why typing far away can feel slow, and a better Telnet could help.