mem usage

[Matthew Fluet]

> I read reports on the various VM problems with 2.4 kernels in Linux and there is
> a lot of variability, especially before 2.4.13.  Also after that there are
> 2 separate vm systems: a brand new one that Linus's kernel uses and the older
> one that Alan Cox is using.  The problem's pre 2.4.13 largely showed up under
> heavy memory load, so that could be part of what you are seeing when 85%
> causes trouble.

I've been reading some of the same things; the problem I mentioned a few
weeks ago with the strange paging behaviour during monomorphisation is
probably a symptom of this.

This time, I've been getting "Out of memory" errors when doing a self
compile of the current CVS sources with the 20011006 release on a machine
with 512M with no compiler switches.  I don't care which VM is running on
that kernel, I shouldn't have a problem getting 85% of the physical RAM.
The fact that I'm seeing out of memory means that the live data is just
too much.  I can sneak by adding @MLton fixed-heap 500m -- to the compiler
flags, but there is lots of paging going on.  That could be VM stuff;
should an unloaded machine be able to get by with just 12M for the kernel?