[olug] Interactive shell cluster

[Daniel Pfile]

Well, if you just want round robin you could set up a second ip on the
head node and have iptables forward 22 to a round robin dns entry that
holds your cluster node ips.

Or you could make a login script that does the same thing, but your
users would have to enter passwords twice, need to use keys, or
whatever. Of course then you could have the script just check a file in
/etc for the current host to go to based on output from your monitoring
software

Or combine a bit and forward 22 to an alias in /etc/hosts or dns that
gets updated somehow.

-- Daniel


Carl Lundstedt wrote:
> Hey Nebraska Luggers,
> 
> I'm taking a shot in the dark here, does anyone know how to configure an
> interactive shell cluster?  
> 
> What I have is about 8 1U servers with 4CPUs each.  One acts as a
> headnode, the others as 'workers' (standard HPC cluster with PBS and
> Condor available for scheduling).  What I'd like is for anyone logging
> into the headnode to spawn a shell on a worker so that interactive CPU
> usage can be balanced.  I'd settle for round robin scheduling, it
> doesn't have to be 'smart' about it.
> 
> I know this is done at Fermilab, but I can't find any good docs anywhere
> on how to configure such a beast.  Perhaps it's really trivial?
> 
> Can someone point me to a good how to?  Google didn't seem to help last
> time I tried it.
> 
> Carl Lundstedt
> UNL
> 
> _______________________________________________
> OLUG mailing list
> OLUG at olug.org
> http://lists.olug.org/mailman/listinfo/olug