[torqueusers] checkjob shows 16-CPU job when extra CPUs were
ataufer at clusterresources.com
Thu May 22 09:38:02 MDT 2008
You might want to try one of the latest Torque snapshots dated May 19,
2008 or later. A new configure option has been added. This option
--enable-maxnotdefault will make it so Torque does not use resources_max
for the queue or server as defaults for jobs.
Peter van Heusden wrote:
> Thanks Al. Seems very non-intuitive though - I thought resources_max
> specifies the *maximum* resources a job can require? And then
> resources_available is how much is available...
> Oh well, I guess I also need to explicitely specify resources_default
> for the queue (or for the server)?
> Al Taufer wrote:
>> If you have set your resources_max.ncpus to 16, then that is probably
>> where the value is coming from.
>> Torque applies values to resource attributes in the following order:
>> 1) Apply values specified on job submission
>> 2) Apply values from queue resources_defaults for attributes that are
>> not yet set
>> 3) Apply values from server resources_defaults for attributes that are
>> yet not set
>> 4) Apply values from queue resources_max for attributes that are not yet
>> 5) Apply values from server resources_max for attributes that are not
>> yet set
>> Peter van Heusden wrote:
>>> I'm running Torque 2.3.0 with Maui 3.2.6 p 19. We have a queue
>>> configured in Torque with resources_max.ncpus and
>>> resources_available.ncpus. A job submitted with a simple qsub -q long
>>> somehow ends up with 16 CPUs dedicated (as shown by maui's checkjob).
>>> Any idea how this happens? Is there some default for the number of CPUs
>>> to use set somewhere?
>>> torqueusers mailing list
>>> torqueusers at supercluster.org
> torqueusers mailing list
> torqueusers at supercluster.org
More information about the torqueusers