[torquedev] Patch: spread job polling more uniformly

Eygene Ryabinkin rea+maui at grid.kiae.ru
Wed Jan 25 23:40:04 MST 2012


Lukasz, good day.

Wed, Jan 25, 2012 at 02:57:53PM +0100, Lukasz Flis wrote:
> How are the tests going on your cluster?

Well, they're good: our Torque-Maui tandem now responds even more
quickly and the percentage for periods of slow responses from Torque
server dropped by approx 7 times.  But our worker nodes are heavily
loaded with I/O-intensive tasks, so we see this improvement.  For more
responsive pbs_moms the problem I am fighting with shouldn't come up.

> I would like to give it a try 12k core cluser here in Cyfronet but i 
> would like to be sure that you observed no regressions since then.

No regressions with something like 10k+ jobs per day.

> Have you measured the improvements over original code?

Yes, as I said periods of irresponsiveness dropped by 7 times at our
workload.

I wonder if people from SuperCluster can review my patch.
-- 
Eygene Ryabinkin, Russian Research Centre "Kurchatov Institute"


More information about the torquedev mailing list