PBS Mom (execution component) should be deployed on both the nodes
Passwordless-ssh should be configured for all the users (StrictHostkeychecking should be “no” in the /etc/ssh/ssh_config on systems in the PBS Cluster)
headnode to compute node(s)
compute node(s) to headnode
compute node to compute node(s)
There should be a common share where users can access from all the systems in the PBS Cluster
For eg: export /workingdirectory from the headnode and mount it across all the compute nodes, makes sure this /workingdirectory is readable writeable by all the users from across all the compute nodes.
Make sure you have a application or the script you are trying to run supports MPI ( Intel, OpenMPI or others ) and is located in the /workingdirectory (along with input files), submit the job script as below qsub -l select=2:ncpus=2:mpiprocs=2 -l place=scatter .
Please share more information on the configuration of your setup and the application or job script that you are trying to run and issues / logs you have come across.
My issue was, I did not have enough resources. I identified when I ran qstat -f.
Could you share the procedure for setting up passwordless ssh between nodes.
I usually add the keys between nodes using ssh-copyid user@node01. But this would be tiresome for huge clusters. I am creating the users on nfs shared location which is accessible by all nodes. Is there any simpler way?