Chris Dagdigian wrote: ... > This is not optimal because the gridMathematica job will still launch > and run outside the control of the cluster scheduler and resource > management layers. On the plus side however we at least allow for the > cluster software to make informed decisions on which nodes the math jobs > run on. This is the scheduler bubble I indicated. It is not so hard to launch an MPI job where all it does is "sleep" in a loop. Then you have your list of machines from the DRM, and can launch in there. The bubble is hard on the DRM as you are effectively requesting a static resource in the middle of your dynamic system. They often do not play well together. From SGE, I use the mpich parallel environment to get the machine file to be autogenerated q{sub,rsh,...} -pe mpich N ... Then you have $TMPDIR/machines available to pull out the machines file. I am sure a similar "trick" exists under LSF (though as I do not have it on my machines at the moment, I cannot tell you what it is). There are other techniques, and just like these others, they are hacks. But some of them work, so those are good.... -- Joseph Landman, Ph.D Scalable Informatics LLC, email: landman@scalableinformatics.com web : http://scalableinformatics.com phone: +1 734 612 4615