[GE users] Configuring a queue to require memory amount?

Reuti reuti at staff.uni-marburg.de
Fri May 6 09:10:04 BST 2005


    [ The following text is in the "ISO-8859-1" character set. ]
    [ Your display is set for the "ISO-8859-10" character set.  ]
    [ Some special characters may be displayed incorrectly. ]

Hi,

you could define a boolean complex like "bigmem" and attach it as "forced" to 
your upgraded machines. Also a memory complex with the reverse relation could 
be used (but you would have to specify in the qsub the two memory values - for 
h_vmem and the minimum). I don't know the rest of your setup, but maybe h_vmem 
isn't necessary any longer with a new introduced complex.

But: depending on your job turnaround, you may face idling nodes with such a 
setup (get many small jobs waiting, and the upgraded machines are doing 
nothing). Maybe it's better to put the upgarded machines in a hostgroup and 
give them a higher sequence number in the queue definition, so that they get 
filled last with small jobs (plus scheduler set queue_sort_method to seqno) 
instead.

CU - Reuti


Quoting "Boone J. Severson" <severson at cray.com>:

> All,
> 
> Looking through qconf it seems h_vmem is a way to specify a per-job 
> maximum limit, but I'm interested in specifying a 
> minimum-requested-memory required resource that will not allow small 
> memory jobs on? We got some memory upgrades to 16 GB but only on a 
> subset of machines, and we'd like to keep those open to the jobs that 
> really need them.
> 
> Thanks,
> Boone
> 
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
> For additional commands, e-mail: users-help at gridengine.sunsource.net
> 



---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
For additional commands, e-mail: users-help at gridengine.sunsource.net




More information about the gridengine-users mailing list