I am pretty sure that I see the problem and it is related to job preemption 
combined with a relatively large number of jobs. I'll see if we can speed
things up.
________________________________________
From: [email protected] [[email protected]] On Behalf 
Of Bjørn-Helge Mevik [[email protected]]
Sent: Wednesday, February 16, 2011 6:50 AM
To: [email protected]
Subject: Re: [slurm-dev] Slow backfill testing of some jobs.

[email protected] (Bjørn-Helge Mevik) writes:

> At the time, about 1000 running jobs, and about 1000 queued jobs.

I forgot: we tried the same jobs again, now with 1189 running jobs (7 of
which are lowpri), 323 pending jobs (no lowpri jobs).

Specifying --ntasks=1 --nodelist=compute-0-1 --constraint=hugemem took 6 
seconds,
specifying --ntasks=1 --nodelist=compute-0-1 --constraint=hugemem*1 took 3 
seconds.

--
Bjørn-Helge Mevik, dr. scient,
Research Computing Services, University of Oslo


Reply via email to