[gridengine users] Blocking all the slots on the node for one job

Bhavishya Goel bhavishya.goel at gmail.com
Wed Mar 27 12:09:15 UTC 2013


So the option of using openmp parallel environment doesn't really work. So
if I have two queues, q1 and q2 for two set of users, and if q1 has
occupied all the slots on one node, the grid engine still schedules the
jobs from q2 on the same node!  So my qstat output may look like this:

queuename qtype resv/used/tot. load_avg arch          states
---------------------------------------------------------------------------------
q1 at node-1 BIP   0/8/8          0.01     linux-x64
---------------------------------------------------------------------------------
q2 at node-1 BIP   0/4/8          0.01     linux-x64

Is there a way to fix this?


On Tue, Mar 26, 2013 at 5:02 PM, Reuti <reuti at staff.uni-marburg.de> wrote:

> Hi,
>
> Am 26.03.2013 um 16:39 schrieb Bhavishya Goel:
>
> > I want to schedule a single job on the cluster using grid engine, but
> while running that job, I want to block all the slots on that particular
> node during the execution of that job. I need to do this for benchmarking
> reasons. The easiest way that I can think of is to use the openmp parallel
> environment with the number of required slots equal to the number of slots
> on one node. That way (as per my understanding) grid engine won't schedule
> my job unless there are 8 slots available on a single node and won't
> allocate the slots on that node to other jobs while job is executing. Is my
> understanding correct? Is there a better/easier way of doing this?
>
> Correct, this is one way of doing it: requesting a PE with
> "allocation_rule $pe_slots" and requesting all slots on a machine which
> still allows you to execute a serial job only.
>
> Another way could be to define a complex as "BOOL EXCL" (`man complex`),
> attach it at least to this particular exechost and request it at the job
> submission. This will ensure that you get this exechost *) for this job
> alone too.
>
> -- Reuti
>
> *) The complex can also be attached to a queue to get a queue instance on
> your own, but this won't help here in case you have more than one queue per
> exechost. Then attaching it to the exechost is the only working way.
>
>
> > --
> > ಠ_ಠ
> > _______________________________________________
> > users mailing list
> > users at gridengine.org
> > https://gridengine.org/mailman/listinfo/users
>
>


-- 
ಠ_ಠ
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://gridengine.org/pipermail/users/attachments/20130327/2cb9aa24/attachment.html>


More information about the users mailing list