[GE users] Background jobs

reuti reuti at staff.uni-marburg.de
Tue Aug 4 23:45:08 BST 2009


Am 04.08.2009 um 21:12 schrieb eddale:

> I've been tasked with creating a grid configuration that saturates the
> grid with background jobs at all times.  Anytime a foreground job is
> submitted, it should take priority and execute.  My boss would like  
> this
> to be done by having an equal number of foreground and background  
> slots
> so that a 16-cpu node would have 32 slots, 16 belonging in the
> foreground queue and 16 in the background queue.  Background jobs  
> should
> be run with a nice value of +20 and never be suspended with the  
> thought
> that the nice value will eliminate any contention.

as long as you have less than 16 jobs running (i.e. slots used) on  
such a machine in total, the nice value doesn't matter and all will  
run at full speed. The nice value is only a relative hint when more  
jobs than builtin cores  are eligible to run, it's not an absolute  
value. Whether the percentage of the two job types satisfies your  
needs, you have to check with some jobs.

Your intended setup is feasible and often done. Only limiting factor  
might be the available RAM in the machine (or scratch space in /tmp  
or alike).

-- Reuti

> Does anybody have any thoughts on this configuration?  I'm not crazy
> about always oversubscribing the nodes.  Ideally, we'd be able to use
> the slot-wise suspension that Harald Pollinger posted previously,  
> but we
> can't wait for that to be implemented.
> Thanks,
> Edward
> ------------------------------------------------------
> http://gridengine.sunsource.net/ds/viewMessage.do? 
> dsForumId=38&dsMessageId=210932
> To unsubscribe from this discussion, e-mail: [users- 
> unsubscribe at gridengine.sunsource.net].


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

More information about the gridengine-users mailing list