FW: [torqueusers] help with use of PVM with Torque

Mikko Huhtala mhuhtala at abo.fi
Thu Dec 1 02:47:04 MST 2005


Chris Samuel writes:
 > On Thursday 01 December 2005 06:48, Daniel.G.Roberts at sanofi-aventis.com wrote:
 > 
 > > The trick is to start the master pvmd on a compute node and then add the
 > > slaves..
 > 
 > The problem is that I believe that it uses rsh/ssh internally to launch those 
 > slaves on other compute nodes and the cluster has rsh/ssh completely 
 > disabled. :-(
 > 
 > I was wondering if you could use mpiexec to launch the pvm daemons, but I 
 > don't know enough about PVM to know whether it could work..
 > 

We got around this problem by allowing rsh without passwords between
the batch nodes, but denying it from anywhere else, including the
interactive nodes. This is easy to do with xinetd and sshd
configuration (root is allowed to ssh everywhere, of course). This
way, the pvm daemons can be started and killed from the job script. As
far as I know, the current version of PVM does not work with anything
else besides rsh.

The problem is that this setup allows the users to bypass the batch
system by writing a job script that uses rsh to launch processes the
batch nodes. We have relatively few users and they can more or less be
trusted, so this setup has worked for us.

Mikko



More information about the torqueusers mailing list