Dynamic ncpus/nodes/ppn specification


#1

Hi,
I am very new to PBS. I have an initial test bed installation to mimic 3 HPCs and a bank of other servers.
I have jobs being submitted with the nodes and ppn specifications, but I want to know if and how I can configure the queues to automatically decide on the amount of nodes to submit too.
E.g. As default I would like the jobs to run on 16 nodes, however if the workload is low, I would like the jobs to use more nodes, say 32, so that it clears the queue faster and doesn’t have resource sitting idle.

Is it possible to do this within PBS? Or would I have to do something like, specify the maximum nodes on the queue and as jobs queue up, I have my own scripts to administer qalter commands to reduce the node count…

Thanks in advance,
MT


#2

PBS does not currently provide the capability to allow users to specify a range of resources to use based on the current usage of the cluster. It has been discussed on the forum under PP-506. Please take a look and add any suggestions you may have.