I built a 2.0.29 SMP kernel, no special patches.
After being up fo about a week I had a total freeze. I tried the
program again and got the lockup every time. No panics. No console
switching. Nothing.
Now, the program uses abour 17MB of memory, most of it very active.
I just made the program use a 16MB (in one malloc) instead of the
1MB it used before. I lowered the size to 8MB and it still locked,
but a bit later. Only when I got down to 4MB did it manage to
complete.
The machine has 64MB memory and about 70MB swap.
To make it clear, with 16MB (a total of 33MB used) the program locks
up at practically the same point (I tested three times).
Root and swap are on IDE, other data on SCSI.
BTW, I had top going during the later tests and there was hardly any
swap used (30-50KB). My program was the only real program used,
no remote or other users. I saw nothing unusual up to the freeze
point, but there was some unusual disk activity at that moment
(well, hard to be sure here as the disk was used constantly) - maybe
the cron job doing its thing for that minute?
I should mention that when it works it does it very well. At one time
I had a CPU intensive program that ran in about 30 minutes. I then
launched it twice (at the same time) and the two jobs took only a few
seconds more (each) than the one done alone. No SMP overhead to speak
of. Real nice.
Can't wait for 2.1 to get somewhat stable and gain finer grain SMP.
-- Eyal Lebedinsky (eyal@eyal.emu.id.au)