Queues
Default queue
Queues are a basic concept of PBS. When submitting a job, the user normally does not have to specify any queue. There exists a default queue, namely default@pbs-m1.metacentrum.cz
where the job goes if not specified otherwise. Then it is automatically sorted to respective queues based on duration of the job (walltime) and/or numbers of GPUs.
Specific queues
In some cases, it is possible or even recommended that the user choses a particular queue. This is the case when the job cannot be routed only according to walltime/no. of GPUs.
Queue name | Description |
---|---|
large_mem@pbs-m1.metacentrum.cz | Jobs requiring 500 GB or more, up to 1 week walltime |
gpu_dgx@pbs-m1.metacentrum.cz | Jobs requiring >= 4 GPUS with NVlink |
uv18@pbs-m1.metacentrum.cz | Jobs requiring >100 CPUs OR >500 GB of memory |
Queue info by qstat
The qstat
command provides info about queues and jobs.
Example:
To see details for a selected queue, use:
Example of output of qstat -Q -f gpu_long@pbs-m1.metacentrum.cz
:
In this particular queue, no jobs can run unless they require at least 1 GPU (resources_min.ngpus
).
Last updated on