2 Replies Latest reply on May 13, 2013 5:08 PM by 1008608

    How to limit the maximum number of jobs grid engine starts

    1008608
      Hello, I'm currently submitting very many jobs to grid engine using qsub and each one of those jobs are very demanding. This sometimes causes the programs to crash with an error that looks something like this "Maximum load threshold exceeded". I want to in some way specify a flag to grid engine that says that grid engine is not to start more than a certain number of jobs at the same time for my user. If I could for instance define that for a certain set of jobs, no more than 3 are to be run at the same time it would solve my problem. It is possible for the admin to define how many jobs a user is allowed to run at the same time, but I would like to specify a maximum for myself which I can easily change depending on the situation.

      Thank you.
        • 1. Re: How to limit the maximum number of jobs grid engine starts
          omarh-oracle - oracle
          Hi,
          Please read the man pages for: sge_conf and sge_resource_quota to get familiar.
          max_u_jobs is a param that may help you.
          1 person found this helpful
          • 2. Re: How to limit the maximum number of jobs grid engine starts
            1008608
            Thank you for your reply it was helpful, but it doesn't quite solve my problem. I want to be able to decide the number of jobs that will start from the qsub queue also I'm not an admin, although I have contact with the admin I need to be able to swap the number of jobs that will run at the same time frequently. This because of the fact that sometimes we are running jobs that will crash if we have more than 2 jobs at the same node and sometimes we can run significantly more jobs at once than that. Also max_u_jobs kills the process and gives an error if we exceed this limit. I simply want to be able to tell grid engine that it should never try to run more jobs than a certain number, an error message once I try to exceed this limit is not exactly what I need.

            Anyway thanks!