The explanations on how to configure NFS are in the Bright Cluster Administrator Manual.
But here’s a recipe for this particular example of moving from a default local /cm/shared that is provided via the head node, to a /cm/shared provided over an external NFS:
Stopping CMDaemon is not needed if there are no services, applications, or environment modules accessing /cm/shared.
This can be checked with an
lsof /cm/shared
which should show which services (for example, slurmctld and slurmdbd if running SLURM) need to be shut down.
Then, copy the data from the head node’s local /cm/shared to the external NFS server. Rsync can do this, with, for example:
# rsync -av /cm/shared/ nashost:/cm/shared/
Next, add an fsmount entry to the Bright head node:
# cmsh% device use master% fsmounts% add /cm/shared% set device nashost:/cm/shared% set filesystem nfs% commit
Then, modify the existing /cm/shared fsmount for the regular node categories so that the regular nodes mount /cm/shared from the external NFS server, and not from the head node:
# cmsh% category use <category-name>% fsmounts% set /cm/shared device nashost:/cm/shared% commit
Finally, all /cm/shared@* entries should be removed from the fsexports entry on the head node. For example:
# cmsh% device use master% fsexports% remove /cm/shared@internalnet% commit
That will remove all /cm/shared entries from /etc/exports on the head node
In the examples above, replace nashost with the IP address or fully-qualified domain name of the dedicated NFS server