Hi Zhi Any luck with the 1.10.1 nightly snapshot? Are these all being run as separate jobs (i.e., each test is executed in series using a separate mpirun)? Ralph
> On Oct 15, 2015, at 11:48 AM, Jeff Squyres (jsquyres) <jsquy...@cisco.com> > wrote: > > Ralph is our runtime guy; he can provide the definitive answers here. > > But first, can you try the latest v1.10.1 nightly snapshot tarball? Some > fixes have gone in since v1.10.0 has been released: > > http://www.open-mpi.org/nightly/v1.10/ > > >> On Oct 15, 2015, at 1:13 PM, Zhi Zhang <zzh...@lenovo.com> wrote: >> >> Hi, >> >> >> >> When running the Intel MPI Benchmark (versions 3.2.3 and 4.1.0.109 were >> tried), running all benchmarks, and looping the test over and over, after a >> while, when the test is exiting (running MPI_Finalize), the test will hang. >> >> >> >> When the processes on the nodes are checked, one node will be running >> IMB-MPI1 still, and that process will be in zombie state. The orted parent >> of that process is still running though. This has been reproduced with >> multiple networks and transports, but most debug on this was done with Intel >> TrueScale IB. >> >> >> >> I can provide more information if needed. Thanks. >> >> >> >> <image001.png> >> >> >> _______________________________________________ >> users mailing list >> us...@open-mpi.org >> Subscription: http://www.open-mpi.org/mailman/listinfo.cgi/users >> Link to this post: >> http://www.open-mpi.org/community/lists/users/2015/10/27871.php > > > -- > Jeff Squyres > jsquy...@cisco.com > For corporate legal information go to: > http://www.cisco.com/web/about/doing_business/legal/cri/ > > _______________________________________________ > users mailing list > us...@open-mpi.org > Subscription: http://www.open-mpi.org/mailman/listinfo.cgi/users > Link to this post: > http://www.open-mpi.org/community/lists/users/2015/10/27875.php
