The answer is NFS. What was the question?

[Felix Lee]

We have a gaggle of semi-autonomous machines: Sun4s, IBM RTs, each with
its own disks.  We use the automounter to mount people's home directories
(among other things).

The problem is when a machine goes away%.  Any process NFS-busy on that
machine's disk gets hosed until the machine comes back.  This is not
unreasonable (although redundant volumes might be nice).

The real problem is that *everyone* who mounts from that machine
eventually gets hosed.  Because getwd() eventually tries to stat() all the
filesystems in /etc/mtab (to recreate /tmp/.getwd).

Not very friendly.

"Remove the /tmp/.getwd efficiency hack" is not sufficient.  You can still
get screwed if you have NFS directories mounted on the same level.

"Don't use NFS" is inconvenient.  I'm not sure I can convince anyone this
is a good idea unless there's an alternative.

Does anyone have any suggestions?  Is there some magic NFS incantation
that will cure the common cold?

NFS seems adequate for client/fileserver types, but seems lacking for
equal peers.

% The automounter sometimes goes away, but that's a different story.
--
Felix Lee	flee at shire.cs.psu.edu	*!psuvax1!shire!flee