[Condor-users] How to force MPI jobs to run in MPI Universe?

["Yufang Zhang" <zhangyufang@xxxxxxxxxx>]

Hi all:

 

I am trying to get MPI jobs running in MPI Universe. I have configured the Condor Pool according to the manual. And I have recompiled my code using mpich.Then I submited the MPI executable to the dedicated scheduler.But the MPI job behaved strangly:they always stay idle. In the StarterLog of the dedicated resources,there are some error messages:

 

6/8 13:23:26 ******************************************************
6/8 13:23:26 ** condor_starter (CONDOR_STARTER) STARTING UP
6/8 13:23:26 ** /usr/local/condor/sbin/condor_starter
6/8 13:23:26 ** $CondorVersion: 6.7.19 May 10 2006 $
6/8 13:23:26 ** $CondorPlatform: I386-LINUX_RH9 $
6/8 13:23:26 ** PID = 31269
6/8 13:23:26 ** Log last touched 6/8 13:23:21
6/8 13:23:26 ******************************************************
6/8 13:23:26 Using config file: /home/condor/condor_config
6/8 13:23:26 Using local config files: /home/condor/condor_config.local
6/8 13:23:26 DaemonCore: Command Socket at <192.168.10.34:47402>
6/8 13:23:26 Done setting resource limits
6/8 13:23:26 Communicating with shadow <192.168.10.34:34310>
6/8 13:23:26 Submitting machine is "gcnode034.cap"
6/8 13:23:26 File transfer completed successfully.
6/8 13:23:27 Starting a MPI universe job with ID: 98.0
6/8 13:23:27 RemoteSpoolDir not found in JobAd.  Aborting.
6/8 13:23:27 ERROR adding environment variable to job6/8 13:23:27 Failed to start job, exiting
6/8 13:23:27 ShutdownFast all jobs.
6/8 13:23:27 **** condor_starter (condor_STARTER) EXITING WITH STATUS 0

 

Have I made some mistake? Can anyone tell me what is wrong about it?

Thank you in advance for your help.

Best wishes. 

 

 

---

Yufang Zhang

2006-06-08