[Zaphod-Users] Myri nodes

[Kai Germaschewski]

On Fri, 14 Oct 2005, Fekete, Balazs M. wrote:

> What is the maximum myri nodes that could be requested? I managed to
> compile and run the river model component of the GFDL Earth System
> Model. When I tried to run it with the usual small number of CPUs I used
> for my test runs, it worked. When I asked for 40 nodes and with 1
> process on each, it failed to establish communications with some CPUs. I
> noticed that the nodes assigned to my task seemed to be mixed (if the
> node name reflect ethernet or myrinet communication). I also figured
> that we had two CPUs on each node, so I should be running 20 nodes with
> two processes on each, so I tried "nodes=20:ppn=2:myri" which hanged
> forever. Perhaps, I should be patient, or simply there is no 20 nodes
> with myrinet.

There's a limit of 64 nodes for a single job. Please use, as you suggested
"ppn=2" instead of requesting just 1 process per node, because in that
case the 2nd processor on all of the nodes your job runs goes unused.

Basically everything you did is fine, you just need patience because there 
aren't enough (20) nodes available for your job at this time, so it'll 
remained queued until its turn comes. "showq" or "qstat" show the 
currently running and queued jobs.

--Kai