Date : Sun, 05 Jun 1983 14:35:00 EDT
From : SJOBRG.ANDY%MIT-OZ@mit-mc
Subject: UNIX on the 68000
I have looked at the specs. I also know how unix works internally. A 512
byte page size is a good size for a virtual memory system. If you make the
page size smaller or larger, you're going to be swapping yourself to
death. If you're not running a virtual memory system, then the problem
of trashing gets worse if you're trying to run (ugh) plain vanilly v7.
If Motorola had done it right with the 68451, they would have put in lots
more pointers, or had just a pointer or two into main memory (albiet slower,
but it provides for much faster context switching.
Unix is a multi-tasking system by definition, and on any systems that have
a number of services availiable (networking, multi-users, watchdogging), you
may have thrashing problems if you don't have moby amounts of memory. You
also need to context-switching from process to process. If it isn't fast
enough, then you get thrashing...
I have at work a little baby 68k system with v7, and there simply isn't
enough processing power to keep >me< happy. There are three problems:
1) It is running v7 ((ugh) swapping instead of winning paging)
2) There simply isn't enough memory. We're running 2meg with 8k
pages, and that isn't enough to pop into vi (or emacs when I
get the tape) from csh - it has to swap out csh in order to
load in vi. This is not good for standard development cycles.
Maybe you're using ed instead?
3) The disk i/o on our little 24meg kennedy winnies is too slow - they
are one of the causes of my thrashing problems (I run about
8 process -- network mailers, and misc daemons). If I had >real<
unix (Berkeley unix), then it would cease to be a problem. Bell
still hasn't put out a working paging unix yet! Even at Bell,
they're running Berkeley on thier vaxen.
In an s100 enviroment, there may be even more bottlenecks than in a multibus
Not only do we need real virtual memory and a good processor for unix to
be happy, but there will have to be faster i/o devices or the damn thing
is going to be i/o bound rather than computation bound (hell, under gosling
emacs, i can keep a 780 computation bound when i'm doing useful work).
Maybe I'll never be happy with a unix micro.