I’ve deployed a PBS cluster with one control node and four compute nodes. Is it mandatory to configure passwordless SSH between these five nodes? What impact will not configuring it have on my usage?
No for PBS , but requires some mechanism to stage-in / stage-out the data.
How about running a MPI jobs involving all these 5 nodes ?
It is a requirement to setup an HPC cluster.
- headnode to compute node(s) and vice versa
- compute node(s) to compute node(s)
Stage-in and Stage-out would be affected (job will be in waiting state - W state)
#PBS -W stagein=... and #PBS -W stageout=... directives
PBS use
###########################################
- rcp - (in the remote past) - It was the default file-transfer mechanism in early legacy versions of PBS ( /etc/pbs.conf , PBS_RCP=/bin/false to avoid it )
- scp - This is the current common standard. When moving data to or from a remote host, the execution daemon (
pbs_mom) calls/bin/scpto securely copy your data (stage-in/stage-out). ( /etc/pbs.conf , PBS_SCP=/bin/scp , by default now ) - cp - This is used if the file path is local to the host or mapped across a shared file network. Check the mom_priv/config for this $usecp directive.
for stage-in and stage-out.