> Am 06.11.2015 um 11:14 schrieb William Hay <w....@ucl.ac.uk>:
> 
> On Fri, Nov 06, 2015 at 02:32:21PM +0800, Wei-Zhao Lu wrote:
>>   HI ALL,
>> 
>>   I have problem about openmpi.
>>   I used openmpi 1.6.5 and ran parallel job normally.
>> 
>>   I upgrade openmpi to 1.8.7 to run the same parallel job failed.
>>   the error message is
>>   --------------------------------------------------------------------------
>>   There are not enough slots available in the system to satisfy the 4 slots
>>   that were requested by the application:
>>     ./cpi
>> 
>>   Either request fewer slots for your application, or make more slots
>>   available
>>   for use.
>>   There are not enough slots available in the system to satisfy the 4 slots
>>   that were requested by the application:
>>     ./cpi
>> 
>>   Either request fewer slots for your application, or make more slots
>>   available
>>   for use.
>>   --------------------------------------------------------------------------
>> 
>>   Because I want each node serve for 1 job, so I configured as follow:
>>   $ qconf -sq all.q | grep slots
>>   slots                 1
>>   How to fix the problem. Thanks.
> 
> My best guess is that your 1.8.7 has grid engine support built in and 
> therefore when run
> under grid engine it only sees as many slots as grid engine allocates.  So 
> that provides
> a couple of options:
> 
> i)Disable grid-engine support if mpirun doesn't provide a switch (I can't 
> recall) then
> you can probably do this by having the job script unset a few of the 
> variables grid engine sets
> 
> ii)Ensure grid engine allocates as many slots as you require.  Change all.q 
> to have
> a number of slots equal to the number of cores on the node (or however many 
> you think
> you will need if that differs for some reason).  Add an exclusive resource 
> (see the complex manpage)
> to each host.   When submitting a job make sure you request a pe with the 
> appropriate number of slots
> and the exclusive resource.
> 
> The second is the cleaner way to do it but does involve changing how people 
> submit jobs.

I definitely recommend the second option too.

To ease the handling for the users, a JSV could add the exclusive resource in 
case a particular PE is requested. It could also adjust the number of requested 
slots from any value to the range "1-" to use the complete node, independent 
from the number of built-in cores which may vary between machines.

-- Reuti


> William
> 
> _______________________________________________
> users mailing list
> users@gridengine.org
> https://gridengine.org/mailman/listinfo/users

Attachment: signature.asc
Description: Message signed with OpenPGP using GPGMail

_______________________________________________
users mailing list
users@gridengine.org
https://gridengine.org/mailman/listinfo/users

Reply via email to