[COFF] "On designing finger-friendly interactions" article at aresluna.org

[Douglas McIlroy via COFF]

Mention of the bad  old days of half-duplex terminals reminded me of how
half-duplex custom distorted capacity estimates for full-duplex Multics.

When Multics first began to work, we at Bell Labs found it could barely
support three users. Meanwhile, MIT was touting it as already a ten-plus
user system, based on extrapolating data from CTSS. There were two major
causes for the discrepancy.

One cause was that program development bogged the system down; the early
PL/I compiler was preternaturally slow. In a vicious cycle, slowness
provided motivation to compile in background, thus putting a double load on
the system.

But slow software was not the only thing that caused Multics to choke. "Think
time" between commands was expected to be the same  on Multics as in CTSS.
In fact, think time could be zero on Multics because full-duplex
communication allowed one to type ahead during program execution, whereas
half-duplex CTSS prevented typing while output from a running command was
possible. Thus a Multics user could burden the system with more interactive
commands per minute than a CTSS user could.

Doug