Hi,
I understood that sge directs the job to a single node. This node is
exclusive to the process. This means that the job is run with 4GB RAM
available.
The same data was processed on a P4 2 GHZ machine with 512 MB of memory
and then the analysis terminated normally without problems (fsl 3.3 fc4 32
bits compilation). I ran the job again on my computer with the same data
on the fsl4 64 bit centos4 build (same as the one on the cluster) and this
also terminated normally.
Unfortunately I cannot (that is, may not) log into the nodes directly and
circumvent the job scheduler.
If it is not RAM, or the job timeout/memory limits imposed by the sge
scheduler, do you have any ideas where to look for an answer. The grid
runs on clustervision OS.
Thank you,
Niels van Strien
|