Thanks to all who replied.
The initial problem was a such:
On a couple machines a really large job was causing swap space below 10%
errors. Couldn't figure out why since there was more than ample physical
RAM available. After running the jobs however it seemed that the physical
RAM was not being released and there was not accounting for the large
amount of used phys RAM
Turns out there were a couple issues:
1) The machines that were having trouble had the odd configuration of
having more phys RAM than virtual RAM.
The ones that seemed to be ok had either more virt RAM than actual or
approx. the same amount of each.
2) All machines were running in 'eager' virt mem allocation mode
Which I suppose is ok as long as virt mem >= phy mem
Solution:
1) increase swap space
2) Turn off 'eager' mode of VM operation.
In Tru64 5.0 (no patches) the way to go about this is in the file
/etc/sysconfigtab - which the man pages suggest not manually editing.
So using the 'kernel tuner', under the 'vm' section, I set the variable
vm-swap-eager = 0 to turn off eager mode. This is documented in the man
page for 'swapon'
This variable is not changable on the fly, so a reboot is needed after you
change the value. At this time, this seems to have fixed the problem.
Additionally, one can examine the output of 'swapon -s' and look for an
entry called 'reserved' as an indication of eager mode.
Thanks again,
Russ
O===========================================O
| R U S S E L L G A U L D |
| +------------------+------------------+ |
| Computational Fluid Dynamics Group |
| Department of Mechanical Engineering |
| P.O. Box 116300 |
| University of Florida |
| Gainesville, FL 32611 |
| 352.392.4442 |
| * * * |
O===========================================O
Received on Tue Feb 01 2000 - 18:38:35 NZDT