Mounted File System: NFS
From Debian Clusters
The Network File System (NFS) allows the entire cluster of computers to share part of their file system - very important for running parallel code like MPI. In this paradigm, one (or more) machines hold the actual files on their physical hard drive and act as an NFS server while the other nodes "mount" the file system locally. To the user, it looks like their files exist on all of the machines at once.
With NFS, though multiple servers can exist, they must be mounted in different places. (Two machines cannot both share their /home directory and have everyone else mount them, but one machine can share /home/students and another can share /home/faculty.)
I'll use an NFS share for all of my users' home directories, as well as shared software.
NFS setup is relatively easy, though it varies by what the machine's role will be.
Troubleshooting
Sometimes an error like the following occurs:
peregrine:~# mount /shared mount: RPC: Timed out
Restarting the portmap service has worked in these situations for me:
peregrine:~# /etc/init.d/portmap stop Stopping portmap daemon.... peregrine:~# /etc/init.d/portmap start Starting portmap daemon.... peregrine:~# mount /shared
Links
- man exports
