On 17.01.2012 15:14, Reuti wrote:
> Hi,
>
> Am 17.01.2012 um 15:02 schrieb Bartosz Biegun:
>
>> I have 6 subclusters (each subcluster has uniform architecture,
>> motherboard, cpu speed, etc, but they vary between subclusters). It is
> so you mean something like hostgroups, and all share one and the same qmaster?
Yes
>> possible to run program A on any of the subclusters.
>> I want to create queue Aq, wich contains all hosts in all subclusters. I
>> want a job to run only on one particular subcluster (to avoid running on
>> computers with different parameters).
>>
>> For example:
>> subcluster    free cpus
>> sub1        2
>> sub2        4
>> sub3        2
>> ...
>>
>> Submitted job requires 4 nodes.
>> I want the queue to assign the job to the subcluster that has 4 free
>> cpus (sub2).
>>
>> I can create as many queues as subclusters (A_sub1_q, A_sub2_q, ...),
>> but choosing subcluster should be transparent for users.
>>
>> I have try setting complex arch on hosts, but sun grid does not seem to
>> take over architecture (job running on different arch, despite it's
>> suitable number of cpus on of hosts with the same arch).
> So, it's about parallel jobs. What you need are various PEs (one for each 
> hostgroup), then you can even stick with one and the same queue.
>
> http://www.gridengine.info/2006/02/14/grouping-jobs-to-nodes-via-wildcard-pes/
>
> In contrast to the information above one queue is sufficient nowadays:
>
> $ qconf -sq all.q
> ...
> pe_list NONE,[@part1=smp1 openmpi1 mpich1],[@part2=smp2 openmpi2 mpich2]
>
> -- Reuti
Thanks a lot!

-- 
Z poważaniem,

Bartosz Biegun
specjalista IT
Grupa Chemia Kwantowa I/Laboratorium Technik Obliczeniowych
Instytut Katalizy i Fizykochemii Powierzchnii PAN 
Kraków
tel. +48 12 639 52 08, fax. +48 12 425 19 23

_______________________________________________
users mailing list
[email protected]
https://gridengine.org/mailman/listinfo/users

Reply via email to