Thread: How to change the dispacher behavior on multiple nodes?

Hi,

we have a cluster with 32 nodes of 4 cpus each. I noted when i submited a parallel job the dispacher executes the job on one cpu of diferent nodes.
For exemple a job of 5 cpus runs on cpu0 of node1, cpu0 of node 2, cpu 1 of node3, cpu 0 of node4 and cpu 1 of node5. Is there a way to change this behavior to run the job o same node and change the node only when its all cpus are used.
The previous job would run on cpu0 of node1, cpu1 of node1, cpu2 of node1, cpu3 of node1 and cpu1 of node2, for example.

If this LSF standalong (none-EGO version), use the -R "span" option in bsub, search bsub manuel, you can not miss it.

If this is LSF on EGO, then you are a bit out of luck, the best bet is to use "Stacked" slot allocation policy (can be configured in resource plan GUI), this way you will EGO will try to allocate a full host to you, but not guaranteed, then you still specify "span" in bsub.