Norman RAM problem
Matthew Fluet
fluet@CS.Cornell.EDU
Thu, 18 Oct 2001 13:02:09 -0400 (EDT)
> I looked at the output of Norman running MLton, and everything looks fine to
> me. The max semi-space size is 225,017,856 and 85% of 512 megabytes over 2
> is 228,170,138 bytes. I guess that he said he was just seeing swapping, but
> that just means that other processes were fighting for the RAM (or else Linux
> isn't swapping enough out). I did some experiments on my machine (.5 gig,
> Kernel 2.4.2) running through huge chunks of memory and seeing when I seemed
> to start to swap. This was X and Netscape running, but not doing much other
> than in a few X-terms. I didn't see any problems.
Just as a side comment, has anyone else noticed excessive swapping on
self-compiles during some of the early stages (monomorphisation, closure
conversion)? If I do a self compile on my home machine, the system crawls
for about 45secs during one of these stages; watching top reveals that
MLton is only using around 65% memory at that point in time (on a .75G
machine), but kswapd is running like crazy. Maybe its just poor virtual
memory management in the kernel, combined with the dynamic heap-sizing in
MLton, that just interacts badly at that point in time when memory usage
starts to climb.