[GE users] array job has taken over

reuti reuti at staff.uni-marburg.de
Wed Aug 12 22:41:08 BST 2009


Am 12.08.2009 um 23:33 schrieb mbay2002:

> reuti wrote:
>> Am 12.08.2009 um 20:04 schrieb mbay2002:
>>
>>> Hi -
>>> I'm pretty new to SGE, I know very basic administrative stuff and am
>>> still learning.
>>>
>>> We have a vanilla install with 144 nodes, each with 8 cores.  We  
>>> have
>>> not done any policy configuration.
>>>
>>> A user has submitted an array job with 36000 elements.  It was the
>>> earliest job submitted, and now her job has taken over every core
>>> on the
>>> cluster.
>>>
>>> We have several users waiting who just need a few cores, some just
>>> need
>>> a single core.  Many of their jobs have a higher priority than the
>>> running array job.
>>>
>>> But, what happens is that everytime the array job index increases,
>>> another one starts to execute as the core is freed up.
>>
>> You can put the job on hold:
>>
>> $ qhold <jobid>
>>
>> the running ones will continue, but no new ones will start.
>>
>
>
> After I posted, I tried
> qalter -h <jobid>
>
> Is this any different than qhold (especially since I noticed that the
> qhold manpage says to use qalter to remove the hold).

It also mentions qrls, which is the primary counterpart IMO. In the  
end: qhold/qrls are symlinks to qalter - just a shortcut.

-- Reuti

>
> This is a good temporary solution, but hopefully I can come up with a
> policy driven solution so that I do not need to babysit the jobs.
>
> ------------------------------------------------------
> http://gridengine.sunsource.net/ds/viewMessage.do? 
> dsForumId=38&dsMessageId=212057
>
> To unsubscribe from this discussion, e-mail: [users- 
> unsubscribe at gridengine.sunsource.net].

------------------------------------------------------
http://gridengine.sunsource.net/ds/viewMessage.do?dsForumId=38&dsMessageId=212059

To unsubscribe from this discussion, e-mail: [users-unsubscribe at gridengine.sunsource.net].



More information about the gridengine-users mailing list