Newsgroups: comp.parallel.mpi From: Tom Goodale Subject: Re: Chain-jobs on T3D Organization: Max-Plank-Institut Fur Gravitationsphysik Date: Wed, 06 May 1998 10:32:36 +0200 Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Message-ID: <35502024.1CFB@aei-potsdam.mpg.de> If you know the amount of time the queue allows you, you could just record the time the process starts, and then when you approach the limit do a checkpoint and submit a new job. Tom tc wrote: > > jjv5@psu.edu wrote: > > Hi, > > that is not completly right. I want to get the remaining time of a running job > within the job. I know how to write scripts, but I need a subroutine or > function, which tells me how many time remains in this queue. I have heard about > an subroutine, but I forgot the name. In my case it can be, that I have to start > a job up to 6 times until I complete one calculation, because the queue length > is 8 hours and the total running time is about 36 to 48 hours. So the problem is > to stop the program at the right time, before the queue stops the job by > itsself. > > Bye TC > > > > > I've tried a few approaches to this with moderate success. I'll assume > > what you are trying to do is get a series of jobs through the queue but > > each job has to wait for the previous one to finish before it starts. The > > problem is you don't want to wait until one job finishes and then submit > > to the queue because you have to wait so long again. > > > > I've added qsub commands to the end of a job script. This causes one > > job to submit the next job as it finishes. This doesn't overcome the > > problem above, but it let's the jobs be submitted as quickly as possible > > without you monitoring it. > > > > -- > Thorsten Coordes > ZARM - University of Bremen Phone + 49 - 421 - 218 - 4330 > Am Fallturm Fax + 49 - 421 - 218 - 2521 > 28359 Bremen Mobil + 49 - 171 - 712 - 5504 > Germany e-Mail tc@zarm.uni-bremen.de