On Tue, Jul 10, 2012 at 4:02 PM, John Young <[email protected]> wrote: >> If you really have a real use-case for setting the # of descriptors in >> the queue config, then let us know and we can implement that in OGS/GE >> (... when time permits). >> > Well... I have an engineer here who want to run a 2048 core job. When > he tries, he gets this error:
I mean, a real use-case that needs the limit to be part of the queue config, and one that can't just use the method I described in: http://gridengine.org/pipermail/users/2012-May/003705.html If you have a real good reason that setting S_DESCRIPTORS & H_DESCRIPTORS in sge_conf(5) is not enough, then we can try to improve things a bit... but the bottom line is that we need a good use-case. (Even if we had unlimited time, we still don't want to add code that duplicates existing code. Having said that, we can remove existing code and switch to the new code if needed.) You should be able to fix the issue by setting S_DESCRIPTORS & H_DESCRIPTORS, or you can check the environment that starts the execd, and see if it has a lower descriptor limit... http://gridscheduler.sourceforge.net/htmlman/htmlman5/sge_conf.html Rayson > > Fatal error in MPI_Send: Other MPI error, error stack: > MPI_Send(173)....................: MPI_Send(buf=0x1edfe380, count=20, > MPI_CHAR, dest=1040, tag=31, comm=0x84000004) failed > MPIDI_EagerContigSend(186).......: failure occurred while attempting to send > an eager message > MPID_nem_tcp_iStartContigMsg(298): > MPID_nem_tcp_connect(849)........: unable to create a socket, Too many open > files (errno 24) > > Fatal error in MPI_Send: Other MPI error, error stack: > MPI_Send(173)....................: MPI_Send(buf=0x2b2304159010, count=20, > MPI_CHAR, dest=1039, tag=31, comm=0x84000004) failed > MPIDI_EagerContigSend(186).......: failure occurred while attempting to send > an eager message > MPID_nem_tcp_iStartContigMsg(298): > MPID_nem_tcp_connect(849)........: unable to create a socket, Too many open > files (errno 24) > > The "Too many open files" error makes me think I need to up > the number of descriptors available, but if there is some other > solution, I'd love to hear it. (This particular engineer's > job runs fine with core numbers up to about 1500-1600, but when > he tries 2048, it bombs with this error every time.) > > JY > > > _______________________________________________ > users mailing list > [email protected] > https://gridengine.org/mailman/listinfo/users _______________________________________________ users mailing list [email protected] https://gridengine.org/mailman/listinfo/users
