Post date: Sep 02, 2013 7:34:18 PM
I noticed that the bgsr jobs that had begun writing (the 10k locus jobs) were entering uninterruptable sleep (unix top state D). This means that they were basically just waiting to write, which slows the runs down considerably. I killed the jobs and tried again with a few modifications. First, I am now writing to /pscratch/A01963476/. This scratch directory is supposed to be set up for large, parallel I/O, we will see. I also requested two processors per node to limit the number of jobs on the nodes that I am running these on (this might help). The new jobs have ids 7974-8003 (there are 30). Many are currently queued, but a few have started. I will check back again later.