Z.113 Requesting NUMA-Aware Resources

This topic only applies for NUMA-aware systems and requires Torque Resource Manager 6.0 and later.

Various NUMA resources can be requested at the time of job submission.

The qsub -L option allows administrators the ability to place jobs at the "task" or "OS process" level to get maximum efficiency out of the available hardware. In addition, multiple, non-symmetric resource requests can be made for the same job using the -L job submission syntax. See 25.2 -L NUMA Resource Request for a complete list of -L values.

For example:

qsub -L tasks=4:lprocs=2:usecores:memory=500mb -L tasks=8:lprocs=4:memory=2gb

Creates two requests. The first will create 4 tasks with two logical processors per task and 500 mb of memory per task. The logical processors will be placed on cores. The second request calls for 8 tasks with 4 logical processors per task and 2 gb of memory per task. The logical processors may be placed on cores or threads since the default placement is allowthreads.

The queue attribute "resources_default" has several options which are not compatible with the qsub -L syntax. If a queue has any of the following "resources_default" options set, the job will be rejected from the queue. "resources_default" options: nodes, size, mppwidth, mem, hostlist, ncpus, procs, pvmem, pmem, vmem, reqattr, software, geometry, opsys, tpn, and trl.

© 2017 Adaptive Computing