Multi-core jobsΒΆ
Submitting a job using multiple cores on the same compute server is done through the -pe
option that allows you to specify the multicores
parallel environment, as well as the desired number of cores:
$ qsub -pe multicores <number_of_cores> -q <QueueName> <script>
For example :
$ qsub -pe multicores 4 -q mc_long test.sh
The queues available for multi-core jobs are the following:
mc_long
mc_huge
mc_longlasting
To know the queues limits please refer to the page Information on scheduling queues.
Attention
Access to all multi-core queues is restricted. You must contact your czar to access this type of resource. Restricted queues FAQ.
The higher the number of requested cores, the more likely the job will wait before executing. It is thus recommended not to ask more than 16 cores. 8 cores is a good compromise between job size and waiting.
Multi-core jobs must be properly configured to use the number of cores requested, accessible through the $NSLOTS
environment variable. The use of more cores than requested may impact the operation of the compute servers and the jobs of other users.
Note
The queues mc_highmem_long
and mc_highmem_huge
give access to a machine dedicated to computations needing a high value of CPU. Please contact our user support to discuss your computing model.
Attention
The memory resource allocated when using the -l
option, such as -l s_rss=4G
, is the memory allocated for one core, and not for the whole job. For the latter, the total memory allocated is then the number of cores requested times the memory resource requested using that option.