By default PBS submits my serial jobs to all the nodes in a queue before using up more resources(cpus) from the nodes.
Can I force PBS to submit my jobs to one node till it exhausts all the CPUS of that node (say 12 cpus; also given that the memory requirement of 12 serial jobs is less than the memory assigned to each node) before it submits the 13th job to the next node.
I want to do this so that later when I want to submit a job with higher memory requirement, I don't go into 'queue' mode because all the nodes have some jobs running.
Ideally I should have separate queues for this purpose but I want my queues to be dynamic in the sense that I may require more large mem jobs which I cant run because though small memory queue is not being used fully, it has some jobs running on all nodes.