Many people install the VDT on NFS. You should be aware that in some circumstances you may see failures unless you put some files onto a local disk.
For each job that runs, Globus keeps a job state file that records information about a job. This is particularly important if a Globus job manager dies: when a new job manager starts up, it can recover the state that was lost.
However, there are times when a new job manager may be restarted because it is believed that the old job manager failed, when it didn't actually fail. When this new job manager starts up, it will immediately exit because the job state file is locked, so it can tell that the old job manager is still running.
In our experience, NFS on Linux is often flakey, particularly the lock server. We have seen times when the lock appears to be active, even though it is not, thus preventing a new job manager from starting up.
If the job state files are not stored on NFS, then this problem can be completely avoided. There are two ways to avoid having these files on NFS:
/var/globus/tmp, but you can choose anything on a local disk
$VDT_LOCATION/globus/etc/globus-job-manager.confand add the following line, substituting the directory you chose:
Future versions of the VDT will help you do this configuration change more automatically and easily.
NOTE: We have had a number of users tell us that they have installed the VDT and Globus on NFS, but never had a problem. This is true. But if you heavily stress your Globus gatekeeper, you have a Linux NFS server, and you experience any failed jobs, this problem could very well crop up and be difficult to diagnose. We strongly recommend putting your job state files on a local disk.