1 Reply Latest reply on Oct 27, 2011 3:54 AM by compres

    Multiprocessing on a core




      isn´t it possible in rckmpi to run more than one process on a core?


      When I try to run more than one process on rck00, I get the following error:

      rank 1 in job 29  rck00_48956   caused collective abort of all ranks

        exit status of rank 1: killed by signal 9


      MPI_init is successfull. This failure appears when the processes try to communicate with each other.


      Thanks in advance

        • 1. Re: Multiprocessing on a core

          This is not possible. 


          The channel protocol of a single process assumes the full MPB is available for itself; if 2 MPI processes are launched in the same core they will interfere.  Also, using other libraries that access the MPB is not possible, for the same reason.


          The MPD process manager tends to oversubscribe the cores when launching multiple MPI jobs.  In order to launch multiple MPI jobs with the MPD, one has to rely in host files.


          - Isaías