[TUHS] Were cron and at done at the same time? Or one before the other?
jnc at mercury.lcs.mit.edu
Thu Dec 10 05:25:36 AEST 2020
> From: Niklas Karlsson
> Just consider Multics, or IBM's "Future System".
Here's a nice irony for you: one of the key people in killing off FS was
reported to me (by someone who should have known) to be Jerry Saltzer (of
Multics fame). That wasn't the only time he did something like thst, either;
when MIT leaned on him to take over Athena, the first thing he did was to take
a lot of their ambitious system plans, and ditch them; they fell back to
mostly 'off the shelf' stuff: pretty much vanilla 4.2, etc.
Multics itself has an interesting story, quite different from the popular
image among those who know little (or nothing) of it. The system, as it was
when Honeywell pulled the plug on further generations of new hardware (in the
mid-80's) was very different from the system as originally envisaged by MIT
(in the mid-60's); it had undergone a massive amount 'experience-based
evolution' during those 20 years.
For instance, the original concept was to have a process per command (like
Unix), but that was dropped early on (because Multics processes were too
'expensive'); they wound up going with doing commands with inter-segment
procedure calls. (Which has the nice benefit that when a command faults, you
can get dropped right into the debugger, with the failed instance there to
look at.) If you read the Organick book on Multics, it describes a much
different system: e.g. in Organick there's a 'linkage segment' (used for
inter-segment pointers, in pure-code segments) per code segment, but in
reality Multics, as distributed, used a single 'combined linkage segment'
(which also contained the stack, also unlike the original design, where the
stack was a separate segment).
There were also numerous places where major sub-systems were re-implemeneted
from scratch, with major changes (often great simplifications): one major
re-do was the New Storage System, but that one had major new features (based
on operationally-shown needs, like the 4.1/.2 Fast File System), so it's not a
'simplification' case. There's one I read about which was much simpler the second
time it was implemented, I think it was IPC?
More information about the TUHS