On Mon, 2008-01-07 at 14:00 -0500, Pann McCuaig wrote:

> 
> A solution we have knocked around is to create a separate "non-compute"
> node for these users, and that node would not NFS-mount /usr/local. The
> users' login shell on the login node would be changed to a script that
> would log them into the "restricted users node," and when they log out
> from that node, they would be logged out of the login node as well.
> 
> Suggestions? Better ideas? Pointers to RTFM? Thanks.

An alternative way to do this would be to install a batch/load levelling
system - such as Gridengine or PBS. Your cluster already has everything
in place for that.
It would be relatively easy then to set up a group of users to be able
to run batch jobs, whilst others don't have the privelege. (In
Gridengine this is done via users/xusers lists).
So all users could have login priveleges to the cluster head node, for
data sharing as you say, but only certain users could launch jobs on the
compute nodes. By the way, if you want to continue to have interactive
use of the nodes, Gridengine allows you to use 'qrsh' to start an
interactive session, just like an ssh session as you currently use.