If you run at 224 and things look okay, then I would suspect something in the upper level switch that spans cabinets. At that point, I'd have to leave it to Mellanox to advise.


On Jun 11, 2013, at 6:55 AM, "Blosch, Edwin L" <edwin.l.blosch@lmco.com> wrote:

I tried adding "-mca btl openib,sm,self"  but it did not make any difference.
 
Jesusí e-mail this morning has got me thinking.  In our system, each cabinet has 224 cores, and we are reaching a different level of the system architecture when we go beyond 224.  I got an additional data point at 256 and found that performance is already falling off. Perhaps I did not build OpenMPI properly to support the Mellanox adapters that are used in the backplane, or I need some configuration setting similar to FAQ #19 in the Tuning/Openfabrics section.
 
From: users-bounces@open-mpi.org [mailto:users-bounces@open-mpi.org] On Behalf Of Ralph Castain
Sent: Sunday, June 09, 2013 6:48 PM
To: Open MPI Users
Subject: Re: [OMPI users] EXTERNAL: Re: Need advice on performance problem
 
Strange - it looks like a classic oversubscription behavior. Another possibility is that it isn't using IB for some reason when extended to the other nodes. What does your cmd line look like? Have you tried adding "-mca btl openib,sm,self" just to ensure it doesn't use TCP for some reason?
 
 
On Jun 9, 2013, at 4:31 PM, "Blosch, Edwin L" <edwin.l.blosch@lmco.com> wrote:


Correct.  20 nodes, 8 cores per dual-socket on each node = 360.
 
From: users-bounces@open-mpi.org [mailto:users-bounces@open-mpi.org] On Behalf Of Ralph Castain
Sent: Sunday, June 09, 2013 6:18 PM
To: Open MPI Users
Subject: Re: [OMPI users] EXTERNAL: Re: Need advice on performance problem
 
So, just to be sure - when you run 320 "cores", you are running across 20 nodes?
 
Just want to ensure we are using "core" the same way - some people confuse cores with hyperthreads.
 
On Jun 9, 2013, at 3:50 PM, "Blosch, Edwin L" <edwin.l.blosch@lmco.com> wrote:



16.  dual-socket Xeon, E5-2670.
 
I am trying a larger model to see if the performance drop-off happens at a different number of cores. 
Also Iím running some intermediate core-count sizes to refine the curve a bit.
I also added mpi_show_mca_params all, and at the same time, btl_openib_use_eager_rdma 1, just to see if that does anything.
 
From: users-bounces@open-mpi.org [mailto:users-bounces@open-mpi.org] On Behalf Of Ralph Castain
Sent: Sunday, June 09, 2013 5:04 PM
To: Open MPI Users
Subject: EXTERNAL: Re: [OMPI users] Need advice on performance problem
 
Looks to me like things are okay thru 160, and then things fall apart after that point. How many cores are on a node?
 
 
On Jun 9, 2013, at 1:59 PM, "Blosch, Edwin L" <edwin.l.blosch@lmco.com> wrote:




Iím having some trouble getting good scaling with OpenMPI 1.6.4 and I donít know where to start looking. This is an Infiniband FDR network with Sandy Bridge nodes.  I am using affinity (--bind-to-core) but no other options. As the number of cores goes up, the message sizes are typically going down. There seem to be lots of options in the FAQ, and I would welcome any advice on where to start.  All these timings are on a completely empty system except for me.
 
Thanks
 
 
    MPI              # cores   Ave. Rate   Std. Dev. %  # timings   Speedup    Efficiency
================================================================================================
MVAPICH            |   16   |    8.6783  |   0.995 % |       2  |   16.000  |  1.0000
MVAPICH            |   48   |    8.7665  |   1.937 % |       3  |   47.517  |  0.9899
MVAPICH            |   80   |    8.8900  |   2.291 % |       3  |   78.095  |  0.9762
MVAPICH            |  160   |    8.9897  |   2.409 % |       3  |  154.457  |  0.9654
MVAPICH            |  320   |    8.9780  |   2.801 % |       3  |  309.317  |  0.9666
MVAPICH            |  480   |    8.9704  |   2.316 % |       3  |  464.366  |  0.9674
MVAPICH            |  640   |    9.0792  |   1.138 % |       3  |  611.739  |  0.9558
MVAPICH            |  720   |    9.1328  |   1.052 % |       3  |  684.162  |  0.9502
MVAPICH            |  800   |    9.1945  |   0.773 % |       3  |  755.079  |  0.9438
OpenMPI            |   16   |    8.6743  |   2.335 % |       2  |   16.000  |  1.0000
OpenMPI            |   48   |    8.7826  |   1.605 % |       2  |   47.408  |  0.9877
OpenMPI            |   80   |    8.8861  |   0.120 % |       2  |   78.093  |  0.9762
OpenMPI            |  160   |    8.9774  |   0.785 % |       2  |  154.598  |  0.9662
OpenMPI            |  320   |   12.0585  |  16.950 % |       2  |  230.191  |  0.7193
OpenMPI            |  480   |   14.8330  |   1.300 % |       2  |  280.701  |  0.5848
OpenMPI            |  640   |   17.1723  |   2.577 % |       3  |  323.283  |  0.5051
OpenMPI            |  720   |   18.2153  |   2.798 % |       3  |  342.868  |  0.4762
OpenMPI            |  800   |   19.3603  |   2.254 % |       3  |  358.434  |  0.4480
_______________________________________________
users mailing list
users@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/users
 
_______________________________________________
users mailing list
users@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/users
 
_______________________________________________
users mailing list
users@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/users
 
_______________________________________________
users mailing list
users@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/users