|
From: Jared F. <jar...@uc...> - 2006-05-17 21:33:23
|
Will do. I don't think my jobs were causing trouble, but I'll switch to the new volume for future jobs. My jobs were reading into memory 1 or 2 files at the beginning and then not touching the disk for hours while the data was processed and then writing output to file. I was only running 2 jobs at a time as well except for last night. Thanks for the info and for moving my jobs to the other queue so that Brian could use the bigmem queue. ----- Original Message ----- From: Allen Day To: Jared Fox ; Brian O'Connor ; nel...@li... ; nel...@li... Sent: Wednesday, May 17, 2006 2:09 PM Subject: cluster tips Hi Jared, When you submit jobs, please specify the queue you with them to go to. Now that we have multiple queues, the default behavior of the qsub utility is to first put jobs to all.q, then to gb4.q, then to celsius.q, faulting down the chain if a queue is currently full. We're working on making all jobs go to only all.q unless otherwise requested, but until then if you do not need the bigmem queue, please explicity request all.q with the "-q all.q" option to qsub. Also, it appears that the I/O your jobs were doing over the last couple of days were putting a heavy load on the xsan NFS processes. It appeared that you were touching a large number of small files. Please copy your files over to nucleus:/clustertmp/jaredfox that are going to be used from the compute nodes. This volume is NFS exported to all the compute nodes. Thanks. -Allen |