RHEL3 poor swapping performance

From: rihad (rihad_at_mail.ru)
Date: 02/27/04 

Date: Fri, 27 Feb 2004 23:21:03 +0400

I have this remote PC server running Red Hat Enterprise Linux ES release 3
(Taroon) (uname -r: 2.4.21-4.0.1.EL), 1GB RAM, Celeron 1.7GHz, hosting several
quite popular web portals, mostly apache1/php+mysql4. The machine has a single
80GB IDE disk, partitioned as follows: several megs on /boot, 2GB for swap and
the rest on / (ext3). Now for the problem: every once in a while, like 5 or 6
times a day, the process count rockets up to 700 or so, with the load average at
around 70-80 and top showing high iowait times, (meaning the machine is
thrashing?) I tried to turn off the swap and for the first time since long the
machine ran without a hitch for more than a day. Then I turned the swap back on,
and it took only a few hours for the machine to fall on its knees and nothing
short of a reboot could bring it back. Then I started experimenting with
different swap sizes (mkswap /dev/hda2 nblocks), but *even* as small a swap area
as 20MB turned out to be dentrimental. Something's seriously broken in this very
RHEL kernel's VM. For now I left the machine with no swap, but I really would
love you gurus to point me at some knobs to play with! No swap is no good, you
know.

TIA