The version of mxm is reports as:  Version     : 1.5.dc8c171

The version of OFED reports as:  MLNX_OFED_LINUX-2.0-2.0.5:

 

Here are some revised scaling numbers after configuring OpenMPI to use MXM.  I’m not sure if I posted medium or small case last time, but this is the “small” case.  By the time you get out to 800 cores, each process talks to between 10 to 16 other processes (this is a physical domain decomposition), and the message sizes can be described by saying there is a distribution from 1K bytes up to 10K bytes (25%), 3 times larger (50%), and 3 times smaller (25%). On the “medium” case, the difference between OpenMPI and MVAPICH is smaller, but OpenMPI is still doing better.

 

 

Scalability - 1 domain per process

   MPI              # cores   Ave. Rate   Std. Dev. %  # timings   Speedup    Efficiency

================================================================================================

MVAPICH            |   16   |    7.5822  |   0.171 % |       3  |   16.000  |  1.0000

MVAPICH            |   48   |    7.7416  |   0.804 % |       3  |   47.011  |  0.9794

MVAPICH            |   80   |    7.6365  |   0.252 % |       3  |   79.431  |  0.9929

MVAPICH            |  160   |    7.4802  |   0.887 % |       3  |  162.182  |  1.0136

MVAPICH            |  256   |    7.7930  |   1.554 % |       3  |  249.073  |  0.9729

MVAPICH            |  320   |    7.7346  |   0.423 % |       3  |  313.695  |  0.9803

MVAPICH            |  480   |    7.9225  |   2.594 % |       3  |  459.378  |  0.9570

MVAPICH            |  640   |    8.3111  |   2.416 % |       3  |  583.866  |  0.9123

MVAPICH            |  800   |    8.9315  |   5.059 % |       3  |  679.137  |  0.8489

OpenMPI            |   16   |    7.5919  |   0.879 % |       3  |   16.000  |  1.0000

OpenMPI            |   48   |    7.7469  |   0.478 % |       3  |   47.040  |  0.9800

OpenMPI            |   80   |    7.6654  |   0.544 % |       3  |   79.233  |  0.9904

OpenMPI            |  160   |    7.7252  |   2.202 % |       3  |  157.239  |  0.9827

OpenMPI            |  256   |    7.7043  |   0.563 % |       3  |  252.265  |  0.9854

OpenMPI            |  320   |    7.6727  |   6.086 % |       3  |  316.629  |  0.9895

OpenMPI            |  480   |    7.7016  |   0.450 % |       3  |  473.163  |  0.9858

OpenMPI            |  640   |    8.0357  |   0.572 % |       3  |  604.651  |  0.9448

OpenMPI            |  800   |    8.4328  |   3.198 % |       3  |  720.223  |  0.9003

 

From: users-bounces@open-mpi.org [mailto:users-bounces@open-mpi.org] On Behalf Of Mike Dubman
Sent: Wednesday, June 12, 2013 7:01 AM
To: Open MPI Users
Subject: Re: [OMPI users] EXTERNAL: Re: Need advice on performance problem

 

Also, what ofed version (ofed_info -s) and mxm version (rpm -qi mxm) do you use?

 

On Wed, Jun 12, 2013 at 3:30 AM, Ralph Castain <rhc@open-mpi.org> wrote:

Great! Would you mind showing the revised table? I'm curious as to the relative performance.



On Jun 11, 2013, at 4:53 PM, eblosch@1scom.net wrote:

> Problem solved. I did not configure with --with-mxm=/opt/mellanox/mcm and
> this location was not auto-detected.  Once I rebuilt with this option,
> everything worked fine. Scaled better than MVAPICH out to 800. MVAPICH
> configure log showed that it had found this component of the OFED stack.
>
> Ed
>
>
>> 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.
>>>
>>> Jesus’ 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 I’m 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:
>>>
>>>
>>>
>>>
>>> I’m having some trouble getting good scaling with OpenMPI 1.6.4 and I
>>> don’t 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
>>
>> _______________________________________________
>> 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