Itay M wrote:
'ncpus' still exists but only in 17 'old' jobs - ones that were
submitted before we made the 'unset' change. I guess I should wait until
these will end and re-test the system?
Possibly, but rather unlikely...
diagnose -n says for example, on node28 :
node28 Busy 0:4 2926:3950 1:1 3871:7641
1.00 DEFAUL [NONE] DEF 2.19 002 [heavy_2:4][light_4:4][b_que
[DEFAULT] [NONE]
WARNING: node 'node28' has more processors utilized than dedicated (4 > 2)
----- --- 6:86 72602:98716 26:26
142420:212774
But this node is running 2 jobs which both does not have 'ncpus'
settings if I use qstat -f on them.
Can you also report the output of checkjob and diagnose -j on these 2
jobs? Do they also have the MEM requirement?
About the MEM requirement: do you mean to unset it to? other than that
we don't use any MEM requierment in our qsub script.
Well, it must be coming from somewhere, quite possibly from a default in
the queue or server configuration. So I'd try unsetting it there.
However, looking at the diagnose -n output above makes me think it is
processor related - judging from the 0:4, for some unknown reason your
jobs consume 2 processors each rather than 1.
Regards,
Jan Ploski
_______________________________________________
mauiusers mailing list
mauiusers@supercluster.org
http://www.supercluster.org/mailman/listinfo/mauiusers
