On Wed, May 2, 2012 at 9:40 PM, Anne Pajon Anne.Pajon@cancer.org.uk wrote:
Hi Peter,
Thanks for your answer.
No the galaxy machine is not a submitting node of the cluster.
That's probably the problem then :)
So I suppose I do have those two options:
(1) making the galaxy server a submitting node to the cluster or (2) install and run galaxy on a cluster node
What would be best? Any suggestions? If (1), any ideas on what needs to be installed?
We went with (1), partly for historical admin reasons of machine ownership - but also this avoided having a single critical point with one machine running both Galaxy and being our cluster head node (since the cluster isn't just used for Galaxy jobs).
Regarding (2), I don't think you want Galaxy running on a cluster compute node - Galaxy isn't that computationally demanding but I wouldn't want the same machine to also be running general cluster jobs. Consider a rogue job submitted to the cluster which consumes too much RAM and brings the node down - that can be annoying, but it would be painful if this also killed your Galaxy server.
It might make sense to put Galaxy on your cluster head node - which might make sharing the data drive simpler too, depending on how your cluster is setup.
Peter