Open MPI logo

Open MPI User's Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Open MPI User's mailing list

Subject: Re: [OMPI users] Working with a CellBlade cluster
From: Mi Yan (miyan_at_[hidden])
Date: 2008-10-29 09:53:59


 "dmesg |grep Node" on Cell will show :
Node 0: CPUS 0-1
Node 1: CPUS 2-3
.
Linux on Cell/BE puts the CPU-node mapping in /sys/devices/system/node
instead of /sys/devices/system/cpu.

Regards,
Mi

                                                                           
             "Lenny
             Verkhovsky"
             <lenny.verkhovsky To
             @gmail.com> "Open MPI Users"
             Sent by: <users_at_[hidden]>
             users-bounces_at_ope cc
             n-mpi.org
                                                                   Subject
                                       Re: [OMPI users] Working with a
             10/27/2008 04:58 CellBlade cluster
             PM
                                                                           
                                                                           
             Please respond to
              Open MPI Users
             <users_at_open-mpi.o
                    rg>
                                                                           
                                                                           

can you update me with the mapping or the way to get it from the OS on the
Cell.

thanks

On Thu, Oct 23, 2008 at 8:08 PM, Mi Yan <miyan_at_[hidden]> wrote:.
  Lenny,

  Thanks.
  I asked the Cell/BE Linux Kernel developer to get the CPU mapping :) The
  mapping is fixed in current kernel.

  Mi
  "Lenny Verkhovsky" <lenny.verkhovsky_at_[hidden]>

                                                                           
                         "Lenny
                         Verkhovsky" <?
                         lenny.verkhovsky
                         @gmail.com>
                         Sent by: To
                         users-bounces_at_op
                         en-mpi.org "Open MPI Users" <
                                                       users_at_[hidden]>
                                                                           
                         10/23/2008 01:52 cc
                         PM
                                                                           
                                                                   Subject
             Please respond to
   Open MPI Users <?,users_at_[hidden]> Re: [OMPI users]
                                                       Working with a
                                                       CellBlade cluster
                                                                           
                                                                           
                                                                           
                                                                           
                                                                           
                                                                           

  According to
  https://svn.open-mpi.org/trac/ompi/milestone/Open%20MPI%201.3 very soon,
  but you can download trunk version http://www.open-mpi.org/svn/ and check
  if it works for you.

  how can you check mapping CPUs by OS , my cat /proc/cpuinfo shows very
  little info
  # cat /proc/cpuinfo
  processor : 0
  cpu : Cell Broadband Engine, altivec supported
  clock : 3200.000000MHz
  revision : 48.0 (pvr 0070 3000)
  processor : 1
  cpu : Cell Broadband Engine, altivec supported
  clock : 3200.000000MHz
  revision : 48.0 (pvr 0070 3000)
  processor : 2
  cpu : Cell Broadband Engine, altivec supported
  clock : 3200.000000MHz
  revision : 48.0 (pvr 0070 3000)
  processor : 3
  cpu : Cell Broadband Engine, altivec supported
  clock : 3200.000000MHz
  revision : 48.0 (pvr 0070 3000)
  timebase : 26666666
  platform : Cell
  machine : CHRP IBM,0793-1RZ

  On Thu, Oct 23, 2008 at 3:00 PM, Mi Yan <miyan_at_[hidden]> wrote:.
        Hi, Lenny,

        So rank file map will be supported in OpenMPI 1.3? I'm using
        OpenMPI1.2.6 and did not find parameter "rmaps_rank_file_".
        Do you have idea when OpenMPI 1.3 will be available? OpenMPI 1.3
        has quite a few features I'm looking for.

        Thanks,

        Mi
        "Lenny Verkhovsky" <lenny.verkhovsky_at_[hidden]>

                                                                           
   "Lenny Verkhovsky" <?
   lenny.verkhovsky_at_[hidden]>
   Sent by: users-bounces_at_[hidden]
                                                                           
                                                                           
   10/23/2008 05:48 AM To
                                                                           
                                                           "Open MPI
                                                           Users" <
                                                           users_at_open-mpi.
                                                           org>
                 Please respond to
       Open MPI Users <?,users_at_[hidden]> cc
                                                                           
                                                                           
                                                                           
                                                                           
                                                                   Subject
                                                                           
                                                           Re: [OMPI
                                                           users] Working
                                                           with a
                                                           CellBlade
                                                           cluster
                                                                           
                                                                           
                                                                           
                                                                           
                                                                           
                                                                           

        Hi,

        If I understand you correctly the most suitable way to do it is by
        paffinity that we have in Open MPI 1.3 and the trank.
        how ever usually OS is distributing processes evenly between
        sockets by it self.

        There still no formal FAQ due to a multiple reasons but you can
        read how to use it in the attached scratch ( there were few name
        changings of the params, so check with ompi_info )

        shared memory is used between processes that share same machine,
        and openib is used between different machines ( hostnames ), no
        special mca params are needed.

        Best Regards
        Lenny,)

        On Sun, Oct 19, 2008 at 10:32 AM, Gilbert Grosdidier <
        grodid_at_[hidden]> wrote:
                    Working with a CellBlade cluster (QS22), the
                    requirement is to have one
                    instance of the executable running on each socket of
                    the blade (there are 2
                    sockets). The application is of the 'domain
                    decomposition' type, and each
                    instance is required to often send/receive data with
                    both the remote blades and
                    the neighbor socket.

                    Question is : which specification must be used for the
                    mca btl component
                    to force 1) shmem type messages when communicating with
                    this neighbor socket,
                    while 2) using openib to communicate with the remote
                    blades ?
                    Is '-mca btl sm,openib,self' suitable for this ?

                    Also, which debug flags could be used to crosscheck
                    that the messages are
                    _actually_ going thru the right channel for a given
                    channel, please ?

                    We are currently using OpenMPI 1.2.5 shipped with
                    RHEL5.2 (ppc64).
                    Which version do you think is currently the most
                    optimised for these
                    processors and problem type ? Should we go towards
                    OpenMPI 1.2.8 instead ?
                    Or even try some OpenMPI 1.3 nightly build ?

                    Thanks in advance for your help, Gilbert.

                    _______________________________________________
                    users mailing list
                    users_at_[hidden]
                    http://www.open-mpi.org/mailman/listinfo.cgi/users

        (See attached file: RANKS_FAQ.doc)
        _______________________________________________

        users mailing list
        users_at_[hidden]
        http://www.open-mpi.org/mailman/listinfo.cgi/users

        _______________________________________________
        users mailing list
        users_at_[hidden]
        http://www.open-mpi.org/mailman/listinfo.cgi/users

  _______________________________________________
  users mailing list
  users_at_[hidden]
  http://www.open-mpi.org/mailman/listinfo.cgi/users

  _______________________________________________
  users mailing list
  users_at_[hidden]
  http://www.open-mpi.org/mailman/listinfo.cgi/users
_______________________________________________
users mailing list
users_at_[hidden]
http://www.open-mpi.org/mailman/listinfo.cgi/users




graycol.gif
pic04795.gif
ecblank.gif