>It does; 16Mb is really too slow. 24Mb is quite usable (although
>linking is still terribly slow) and 32 is pretty good. The
>differences between each of these points are dramatic. The difference
>between 32Mb and 48Mb in terms of linker speed are similarly dramatic
>-- after that more memory doesn't gain you much (unless it's a
>multiuser system of course).
We were doing some benchmarking of a CAD package under X on a 220.
When we realized that the systems we'd proposed were 16MB, and the
system we were testing on was 24MB, we quickly pulled out 8MB RAM.
The CAD package ran about 4 pct FASTER at 16MB. (no paging).
That doesn't mean, of course, that I'd recommend it for a professional
software developer's system... 32MB is good for that.
>>I remember when IBM first announced the RS/6000 -- the minimum
>>configuration was 8 MB. That didn't last long.
>So far as I know they never shipped any 8Mb systems to real customers
>-- the last time I checked AIX wouldn't boot on an 8Mb system (I think
>I last checked at around 3.1.2 -- I'm absolutely certain that 16Mb was
>minimum on 8943 through 9021).
We DID ship a considerable number of 8MB systems to customers... some even
like them! At one time I had a large lab of student workstations with 320's
at 8MB running Professional CADAM under X. Waiting for X to initialize was
a bit painful, but the systems actually ran OK once X and CADAM were loaded.
AIX, both 3.1.x and 3.2, WILL boot on 8MB systems, though you have to build
the kernel differently. If you build it on an 8MB system this is done
automagically. If you're planning to stick a system in the corner to run
a small to medium size FORTRAN program continuously, an 8MB 320 with 160MB
disk may still be one of the cheapest sources of MFLOPS around!
Other than that, though, 16MB is a reasonable minimum, at least for students.