Howard, I searched through the https://github.com/open-mpi/ompi/wiki and didn't find any recent documentation on the master/v2.x series Cray support. Can you give me some pointers? I am trying to get MPI_Comm_spawn() support on a Cray XC40 via using Open MPI's master branch.
On Fri, Jun 26, 2015 at 12:00 AM, Howard Pritchard <hpprit...@gmail.com> wrote: > Hi Nick, > > I will endeavor to put together a wiki for the master/v2.x series specific > to Cray systems > (sans those customers who choose to neither 1) use Cray supported eslogin > setup nor 2) permit users to directly log in to and build apps on service > nodes) that explains best practices for > using Open MPI on Cray XE/XK/XC systems. > > A significant amount of work went in to master, and now the v2.x release > stream to rationalize support for Open MPI on Cray XE/XK/XC systems using > either aprun > or native slurm launch. > > General advice for all on this mailing list, do not use the Open MPI 1.8.X > release > series with direct ugni access enabled on Cray XE/XK/XC . Rather use > master, or as soon as > a release is available, from v2.x. Note that if you are using CCM, the > performance > of Open MPI 1.8.X over the Cray IAA (simulated ibverbs) is pretty good. I > suggest this > as the preferred route for using the 1.8.X release stream on Cray XE/XK/XC. > > Howard > > > 2015-06-25 19:35 GMT-06:00 Nick Radcliffe <nradc...@cray.com>: >> >> Thanks Howard, using master worked for me. >> >> Nick Radcliffe >> Software Engineer >> Cray, Inc. >> ________________________________ >> From: users [users-boun...@open-mpi.org] on behalf of Howard Pritchard >> [hpprit...@gmail.com] >> Sent: Thursday, June 25, 2015 5:11 PM >> To: Open MPI Users >> Subject: Re: [OMPI users] Running with native ugni on a Cray XC >> >> Hi Nick >> >> use master not 1.8.x. for cray xc. also for config do not pay attention >> to cray/lanl platform files. just do config. also if using nativized slurm >> launch with srun not mpirun. >> >> howard >> >> ---------- >> >> sent from my smart phonr so no good type. >> >> Howard >> >> On Jun 25, 2015 2:56 PM, "Nick Radcliffe" <nradc...@cray.com> wrote: >>> >>> Hi, >>> >>> I'm trying to build and run Open MPI 1.8.5 with native ugni on a Cray XC. >>> The build works, but I'm getting this error when I run: >>> >>> nradclif@kay:/lus/scratch/nradclif> aprun -n 2 -N 1 ./osu_latency >>> [nid00014:28784] [db_pmi.c:174:pmi_commit_packed] PMI_KVS_Put: Operation >>> failed >>> [nid00014:28784] [db_pmi.c:457:commit] PMI_KVS_Commit: Operation failed >>> [nid00012:12788] [db_pmi.c:174:pmi_commit_packed] PMI_KVS_Put: Operation >>> failed >>> [nid00012:12788] [db_pmi.c:457:commit] PMI_KVS_Commit: Operation failed >>> # OSU MPI Latency Test >>> # Size Latency (us) >>> osu_latency: btl_ugni_endpoint.c:87: mca_btl_ugni_ep_connect_start: >>> Assertion `0' failed. >>> [nid00012:12788] *** Process received signal *** >>> [nid00012:12788] Signal: Aborted (6) >>> [nid00012:12788] Signal code: (-6) >>> [nid00012:12788] [ 0] /lib64/libpthread.so.0(+0xf850)[0x2aaaab42b850] >>> [nid00012:12788] [ 1] /lib64/libc.so.6(gsignal+0x35)[0x2aaaab66b885] >>> [nid00012:12788] [ 2] /lib64/libc.so.6(abort+0x181)[0x2aaaab66ce61] >>> [nid00012:12788] [ 3] >>> /lib64/libc.so.6(__assert_fail+0xf0)[0x2aaaab664740] >>> [nid00012:12788] [ 4] >>> /lus/scratch/nradclif/openmpi_install/lib/libmpi.so.1(mca_btl_ugni_ep_connect_progress+0x6c9)[0x2aaaaaff9869] >>> [nid00012:12788] [ 5] >>> /lus/scratch/nradclif/openmpi_install/lib/libmpi.so.1(+0x5ae32)[0x2aaaaaf46e32] >>> [nid00012:12788] [ 6] >>> /lus/scratch/nradclif/openmpi_install/lib/libmpi.so.1(mca_btl_ugni_sendi+0x8bd)[0x2aaaaaffaf7d] >>> [nid00012:12788] [ 7] >>> /lus/scratch/nradclif/openmpi_install/lib/libmpi.so.1(+0x1f0c17)[0x2aaaab0dcc17] >>> [nid00012:12788] [ 8] >>> /lus/scratch/nradclif/openmpi_install/lib/libmpi.so.1(mca_pml_ob1_isend+0xa8)[0x2aaaab0dd488] >>> [nid00012:12788] [ 9] >>> /lus/scratch/nradclif/openmpi_install/lib/libmpi.so.1(ompi_coll_tuned_barrier_intra_two_procs+0x11b)[0x2aaaab07e84b] >>> [nid00012:12788] [10] >>> /lus/scratch/nradclif/openmpi_install/lib/libmpi.so.1(PMPI_Barrier+0xb6)[0x2aaaaaf8a7c6] >>> [nid00012:12788] [11] ./osu_latency[0x401114] >>> [nid00012:12788] [12] >>> /lib64/libc.so.6(__libc_start_main+0xe6)[0x2aaaab657c36] >>> [nid00012:12788] [13] ./osu_latency[0x400dd9] >>> [nid00012:12788] *** End of error message *** >>> osu_latency: btl_ugni_endpoint.c:87: mca_btl_ugni_ep_connect_start: >>> Assertion `0' failed. >>> >>> >>> Here's how I build: >>> >>> export FC=ftn (I'm not using Fortran, but the configure fails if >>> it can't find a Fortran compiler) >>> ./configure --prefix=/lus/scratch/nradclif/openmpi_install >>> --enable-mpi-fortran=none >>> --with-platform=contrib/platform/lanl/cray_xe6/debug-lustre >>> make install >>> >>> I didn't modify the debug-lustre file, but I did change cray-common to >>> remove the hard-coding, e.g., rather than using the gemini-specific path >>> "with_pmi=/opt/cray/pmi/2.1.4-1.0000.8596.8.9.gem", I used >>> "with_pmi=/opt/cray/pmi/default". >>> >>> I've tried running different executables with different numbers of >>> ranks/nodes, but they all seem to run into problems with PMI_KVS_Put. >>> >>> Any ideas what could be going wrong? >>> >>> Thanks for any help, >>> Nick >>> _______________________________________________ >>> 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/2015/06/27197.php >> >> >> _______________________________________________ >> 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/2015/06/27199.php > > > > _______________________________________________ > 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/2015/06/27200.php -- Tim Mattox, Ph.D. - tmat...@gmail.com