[GE issues] [Issue 3077] New - master task of larger parallel job might exceed h_vmem limit

pollinger harald.pollinger at sun.com
Tue Jul 7 20:01:03 BST 2009


http://gridengine.sunsource.net/issues/show_bug.cgi?id=3077
                 Issue #|3077
                 Summary|master task of larger parallel job might exceed h_vmem
                        | limit
               Component|gridengine
                 Version|6.2
                Platform|All
                     URL|
              OS/Version|All
                  Status|NEW
       Status whiteboard|
                Keywords|
              Resolution|
              Issue type|DEFECT
                Priority|P4
            Subcomponent|execution
             Assigned to|pollinger
             Reported by|pollinger






------- Additional comments from pollinger at sunsource.net Tue Jul  7 12:01:02 -0700 2009 -------
The master task of a larger parallel job might exceed h_vmem by just starting dozens or hundreds of qrsh clients. The job is then simply
killed, which is unexpected and annyoing for the user.

One might argue that a job that is restricted to some memory usage just can't start more then a specific number of slave tasks, but then SGE
could handle this situation nicer.

One could also argue that starting the slave tasks is part of Grid Engine, not part of the job, so it shouldn't be counted to the h_vmem
limit of the job.

------------------------------------------------------
http://gridengine.sunsource.net/ds/viewMessage.do?dsForumId=36&dsMessageId=206063

To unsubscribe from this discussion, e-mail: [issues-unsubscribe at gridengine.sunsource.net].



More information about the gridengine-users mailing list