Re: [OMPI users] OpenMPI 1.8.6, CentOS 6.3, too many slots = crash

2015-07-24 Thread Ralph Castain
Woohoo!!! Great to hear! If you want the best performance, then yes - you want the NUMA packages installed. Reason: without them, we can’t bind the memory to the same location as the process, and so you will likely see degraded performance. Relates to our other email thread earlier today. > O

Re: [OMPI users] OpenMPI 1.8.6, CentOS 6.3, too many slots = crash

2015-07-24 Thread Lane, William
Ralph, I re-compiled my test code and ran it on almost our entire cluster (excepting the new hyperthread capable blade chassis), anyway, the code ran successfully: mpirun -np 185 -display-devel-map --prefix /hpc/apps/mpi/openmpi/1.8.7/ --hostfile hostfile-noslots --mca btl ^sm --hetero-nodes --b

[OMPI users] Building OpenMPI 1.8.7 on XC30

2015-07-24 Thread Erik Schnetter
I want to build OpenMPI 1.8.7 on a Cray XC30 (Edison at NERSC). I've tried various configuration options, but I am always encountering either OpenMPI build errors, application build errors, or run-time errors. I'm currently looking at < http://www.open-mpi.org/community/lists/users/2015/06/27230.p

Re: [OMPI users] NUMA: Non-local memory access and performance effects on OpenMPI

2015-07-24 Thread Ralph Castain
Hi Bill You actually can get OMPI to split a process across sockets. Let’s say there are 4 cores/socket and 2 sockets/node. You could run two procs on the same node, one split across sockets, by: mpirun -n 1 —map-by core:pe=5 ./app : -n 1 —map-by core:pe=3 ./app The first proc will run on all

[OMPI users] NUMA: Non-local memory access and performance effects on OpenMPI

2015-07-24 Thread Lane, William
I'm just curious, if we run an OpenMPI job and it makes use of non-local memory (i.e. memory tied to another socket) what kind of effects are seen on performance? How would you go about testing the above? I can't think of any command line parameter that would allow one to split an OpenMPI proces

Re: [OMPI users] shared memory performance

2015-07-24 Thread Gilles Gouaillardet
Cristian, one more thing... two containers on the same host cannot communicate with the sm btl. you might want to mpirun with --mca btl tcp,self on one physical machine without container, in order to asses the performance degradation due to using tcp btl and without any containerization effect. C

Re: [OMPI users] shared memory performance

2015-07-24 Thread Harald Servat
Dear Cristian, according to your configuration: a) - 8 Linux containers on the same machine configured with 2 cores b) - 8 physical machines c) - 1 physical machine a) and c) have exactly the same physical computational resources despite the fact that a) is being virtualized and the

Re: [OMPI users] SGE segfaulting with OpenMPI 1.8.6

2015-07-24 Thread m.delorme
Hi, Thanks a lot, it seems to be working using the --prefix on mpirun. I have trouble understanding why using the right flags or exporting "by hand" with -x PATH -x LD_LIBRARY_PATH do not work. In any case, the --prefix option works so that's a good starting point. Thank you again, MD __