From: Axel Kohlmeyer (akohlmey_at_gmail.com)
Date: Mon Oct 05 2009 - 02:14:41 CDT
On Sun, 2009-10-04 at 19:59 -0400, daniel aguayo wrote:
> If my server administrator read this I will have to delete all my
> I usually do that to run multiple steps simulations without wait a
> long time on the queue.
if your sysadmin forces you to run jobs in the least
efficient way, then hit him/her with a giant cluebat.
a smart sysadmin will always encourage people to work
more effectively. however, on our machines we have established
a queue policy based on the _integral_ job time, i.e.
your maximum allowed share of the machine is the same
regardless of how many slots a job requires. or in other
words, if you ask for double the number of nodes (regardless
of whether they are in one of two job scripts), you
have to half the wall clock.
FWIW, this is established by way of mutual agreement with
a little script that computes the total integrated walltime
of all runnable jobs, so people can at all times see who
is the "machine hog".
that kind of policy seems to have worked very well for us
and usually resulted in frequent >90% machine utilization
for clusters with 128 to 600 cpu cores.
-- Dr. Axel Kohlmeyer akohlmey_at_gmail.com Institute for Computational Molecular Science College of Science and Technology Temple University, Philadelphia PA, USA.
This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:53:20 CST