Closed max810 closed 5 years ago
If your computation somehow can use all cores on a node, then you can by either starting dispynode with -n 1
(so all computations only use one processor on a node) or by giving parameter nodes
to JobCluster
with instances of NodeAllocate
where cpus
is given as 1, or instances of subclass of NodeAllocate
overriding allocate
method that returns 1, etc.
Thanks, with cpus=1
eveything seems to work as intended.
Is there a way to launch ONLY ONE job simultaneously per node while using all the CPUs/cores on the node?
So, when on the node the first job is launched, it uses all the available CPUs/cores, it finishes, returns result and ONLY THEN the next job is started on that node.