Or studs, if you prefer.
I'm dealing with an AIX 4.1.5 system (can't be upgraded higher than this;
it's
unsupported hardware). This is *nearly* on topic, the box is circa 1996-7 (an
Apple Network Server 500). It has 512MB of RAM and a 200MHz 604e CPU board,
and a single 18.4GB SCSI-2 drive (non-RAID).
It hummed along very happily until recently when it started hanging up
during backup jobs. Watching it run with vmstat, I noticed that after the
free page count gets under minfree a couple times (eight seems about
average), it hangs up. lsps -a just before the freeze shows barely a couple
percent of the paging space being used, and the avm usage count in vmstat
is hardly out of the ordinary for this system, so I don't think it's running
out of paging space. Increasing it 1.5x didn't make any difference anyway.
Turning down maxperm to reduce the number of file pages allocated by the
AIX VMM didn't make any difference either (I presume the problem is file
pages since the backup job is basically a tar over the network).
Right now I'm solving the problem by setting maxfree stupidly high so that
whenever the free count gets under minfree, the VMM will release a massive
bolus of free RAM pages (presumably from the file cache pool) and thus
keep the system as far away from hitting minfree again for as long as
possible. However, this is technically disgusting and can't be a good idea
for filesystem performance.
Hardware checks out and RAM checks out (both the power-on "LONG RAM TEST" --
and it sure is long -- and the Apple-supplied diagnostic disk).
What can I do to improve this? Is this a bug in 4.1.5?
--
---------------------------------- personal:
http://www.armory.com/~spectre/ --
Cameron Kaiser, Floodgap Systems Ltd * So. Calif., USA * ckaiser at
floodgap.com
-- Aibohphobia, the fear of palindromes. -- Brian Braunschweiger --------------