From owner-chemistry@ccl.net Mon Nov 12 09:11:01 2012 From: "Felix Gut fgut|-|uni-marburg.de" To: CCL Subject: CCL: Queuing system recommendations? Message-Id: <-47876-121112062508-14195-VufT4urpfczE+bnKnxD02w[]server.ccl.net> X-Original-From: Felix Gut Content-Transfer-Encoding: 8bit Content-Type: text/plain; charset=us-ascii Date: Mon, 12 Nov 2012 10:16:29 +0100 Mime-Version: 1.0 (Apple Message framework v1085) Sent to CCL by: Felix Gut [fgut-,-uni-marburg.de] Dear Victor, we use Condor in our group in combination with NFS and have good experience with that. With some experience it can also be used to calculate parallelised jobs as we did with Amber before we got our GPU based system. It is very easy to submit jobs to Condor with specific requirements which nodes can be used based on architecture, memory, node name (licenses) and so on. Best, Felix -- Felix Gut Institute of Pharmaceutical Chemistry Philipps University, Marburg Department of Pharmacy +49(0)6421/28-21339 fgut]*[staff.uni-marburg.de Marbacher Weg 6 35032 Marburg Germany http://agklebe.de/ Am 31.10.2012 um 09:47 schrieb Mikolaj Feliks mikolaj.feliks]|[gmail.com: > > Sent to CCL by: Mikolaj Feliks [mikolaj.feliks(!)gmail.com] > > Hi Victor, > > I`d recommend Slurm. We use it in our lab with great success, for ORCA jobs too. > > https://computing.llnl.gov/linux/slurm/ > > I find it also easier to setup than Torque. > > > Regards, > > Mikolaj > > > On 10/31/2012 12:31 AM, Victor Rosas Garcia rosas.victor#%#gmail.com wrote: >> Hello everybody, >> >> I'm thinking about setting up a queuing system for my cluster, but I am >> really at a loss about how to choose one. I have heard about Torque and >> SQS, but never used any of them. My cluster is a small, Debian-based >> 32-cpu cluster with local scratch disk space for each 4-cpu box. So, I >> would like to know, what do CCL'ers actually use to queue their jobs? >> >> My calcs use a mix of mostly GAMESS-US and GROMACS with some ORCA. Both >> GAMESS and GROMACS can easily use the whole cluster, but the ORCA calcs >> cannot use the whole cluster, thus the need to have them waiting for the >> next available box. >> >> Best regards >> >> Victorhttp://www.ccl.net/chemistry/sub_unsub.shtml> >