Hi Team,
what is the difference between select and nodes.
In an HPC environment, when we are using the " select=1:ncpus=2" jobs are entering in queue states And reason is " Not enough free nodes available"
But, when we replace the resource with “nodes=1:ppn=2” jobs are running fine.
Can you please provide information on this behavior.
Thank You
Atul Yadav
Hello Atul, those jobs should be almost exactly the same as far as the scheduler is concerned. The -lnodes syntax gets translated by PBS into select (and place) statement equivalent. The job’s schedselect attribute is what the scheduler actually looks at when considering a job to run, and the -lnodes=1:ppn=2" job should have ended up as “schedselect = 1:ncpus=2:mpiprocs=2”. The mpiprocs resource is not actually scheduled upon (it just impacts how the $PBS_NODEFILE gets written), so it is really should be the same as far as the scheduler is concerned.
The jobs would also differ in their place resource (-lselect would have defaulted to free, -lnodes to scatter), but since these are just single chunk jobs that should not have mattered.
Please share the full “qstat -f” output for one of each kind of the jobs you mention above, with the command run as root on the PBS server host (or as a different PBS manager, since schedselect does not show up to normal users in qstat -f output).
Thanks.