John,
BTL refers to Byte Transfer Layer, a framework to send/receive point to point messages on different network. It has several components (implementations) like openib, tcp, mx, shared mem, etc.
^openib means "not" to use openib component for p2p messages.
On a side note, do you have an RDMA supporting device ( Infiniband/RoCE/iWarp) ? If so, is OFED installed correctly and is running?
If you do not have, is the OFED running, which it should not, otherwise ?
The message that you are getting could be because of this. As a consequence, if you have a RDMA supported device, you might be getting poor performance.
A wealth of information is available in the FAQ section regarding these things.
--
Sent from my iPhone
On Sep 15, 2012, at 9:49 PM, John Chludzinski <john.chludzinski_at_[hidden]> wrote:
> BTW, I looked up the -mca option:
>
> -mca |--mca <arg0> <arg1>
> Pass context-specific MCA parameters; they are
> considered global if --gmca is not used and only
> one context is specified (arg0 is the parameter
> name; arg1 is the parameter value)
>
> Could you explain the args: btl and ^openib ?
>
> ---John
>
>
> On Sun, Sep 16, 2012 at 12:26 AM, John Chludzinski <john.chludzinski_at_[hidden]> wrote:
> BINGO! That did it. Thanks. ---John
>
>
> On Sat, Sep 15, 2012 at 9:32 PM, Ralph Castain <rhc_at_[hidden]> wrote:
> No - the mca param has to be specified *before* your executable
>
> mpiexec -mca btl ^openib -n 4 ./a.out
>
> Also, note the space between "btl" and "^openib"
>
>
> On Sep 15, 2012, at 5:45 PM, John Chludzinski <john.chludzinski_at_[hidden]> wrote:
>
>> Is this what you intended(?):
>>
>> $ mpiexec -n 4 ./a.out -mca btl^openib
>>
>> librdmacm: couldn't read ABI version.
>> librdmacm: assuming: 4
>> CMA: unable to get RDMA device list
>> --------------------------------------------------------------------------
>> [[5991,1],0]: A high-performance Open MPI point-to-point messaging module
>> was unable to find any relevant network interfaces:
>>
>> Module: OpenFabrics (openib)
>> Host: elzbieta
>>
>> Another transport will be used instead, although this may result in
>> lower performance.
>> --------------------------------------------------------------------------
>> librdmacm: couldn't read ABI version.
>> librdmacm: assuming: 4
>> CMA: unable to get RDMA device list
>> librdmacm: couldn't read ABI version.
>> librdmacm: assuming: 4
>> CMA: unable to get RDMA device list
>> librdmacm: couldn't read ABI version.
>> librdmacm: assuming: 4
>> CMA: unable to get RDMA device list
>> rank= 1 Results: 5.0000000 6.0000000 7.0000000 8.0000000
>> rank= 0 Results: 1.0000000 2.0000000 3.0000000 4.0000000
>> rank= 2 Results: 9.0000000 10.000000 11.000000 12.000000
>> rank= 3 Results: 13.000000 14.000000 15.000000 16.000000
>> [elzbieta:02374] 3 more processes have sent help message help-mpi-btl-base.txt / btl:no-nics
>> [elzbieta:02374] Set MCA parameter "orte_base_help_aggregate" to 0 to see all help / error messages
>>
>>
>> On Sat, Sep 15, 2012 at 8:22 PM, Ralph Castain <rhc_at_[hidden]> wrote:
>> Try adding "-mca btl ^openib" to your cmd line and see if that cleans it up.
>>
>>
>> On Sep 15, 2012, at 12:44 PM, John Chludzinski <john.chludzinski_at_[hidden]> wrote:
>>
>>> There was a bug in the code. So now I get this, which is correct but how do I get rid of all these ABI, CMA, etc. messages?
>>>
>>> $ mpiexec -n 4 ./a.out
>>> librdmacm: couldn't read ABI version.
>>> librdmacm: couldn't read ABI version.
>>> librdmacm: assuming: 4
>>> CMA: unable to get RDMA device list
>>> librdmacm: assuming: 4
>>> CMA: unable to get RDMA device list
>>> CMA: unable to get RDMA device list
>>> librdmacm: couldn't read ABI version.
>>> librdmacm: assuming: 4
>>> librdmacm: couldn't read ABI version.
>>> librdmacm: assuming: 4
>>> CMA: unable to get RDMA device list
>>> --------------------------------------------------------------------------
>>> [[6110,1],1]: A high-performance Open MPI point-to-point messaging module
>>> was unable to find any relevant network interfaces:
>>>
>>> Module: OpenFabrics (openib)
>>> Host: elzbieta
>>>
>>> Another transport will be used instead, although this may result in
>>> lower performance.
>>> --------------------------------------------------------------------------
>>> rank= 1 Results: 5.0000000 6.0000000 7.0000000 8.0000000
>>> rank= 2 Results: 9.0000000 10.000000 11.000000 12.000000
>>> rank= 0 Results: 1.0000000 2.0000000 3.0000000 4.0000000
>>> rank= 3 Results: 13.000000 14.000000 15.000000 16.000000
>>> [elzbieta:02559] 3 more processes have sent help message help-mpi-btl-base.txt / btl:no-nics
>>> [elzbieta:02559] Set MCA parameter "orte_base_help_aggregate" to 0 to see all help / error messages
>>>
>>>
>>> On Sat, Sep 15, 2012 at 3:34 PM, John Chludzinski <john.chludzinski_at_[hidden]> wrote:
>>> BTW, here the example code:
>>>
>>> program scatter
>>> include 'mpif.h'
>>>
>>> integer, parameter :: SIZE=4
>>> integer :: numtasks, rank, sendcount, recvcount, source, ierr
>>> real :: sendbuf(SIZE,SIZE), recvbuf(SIZE)
>>>
>>> ! Fortran stores this array in column major order, so the
>>> ! scatter will actually scatter columns, not rows.
>>> data sendbuf /1.0, 2.0, 3.0, 4.0, &
>>> 5.0, 6.0, 7.0, 8.0, &
>>> 9.0, 10.0, 11.0, 12.0, &
>>> 13.0, 14.0, 15.0, 16.0 /
>>>
>>> call MPI_INIT(ierr)
>>> call MPI_COMM_RANK(MPI_COMM_WORLD, rank, ierr)
>>> call MPI_COMM_SIZE(MPI_COMM_WORLD, numtasks, ierr)
>>>
>>> if (numtasks .eq. SIZE) then
>>> source = 1
>>> sendcount = SIZE
>>> recvcount = SIZE
>>> call MPI_SCATTER(sendbuf, sendcount, MPI_REAL, recvbuf, &
>>> recvcount, MPI_REAL, source, MPI_COMM_WORLD, ierr)
>>> print *, 'rank= ',rank,' Results: ',recvbuf
>>> else
>>> print *, 'Must specify',SIZE,' processors. Terminating.'
>>> endif
>>>
>>> call MPI_FINALIZE(ierr)
>>>
>>> end program
>>>
>>>
>>> On Sat, Sep 15, 2012 at 3:02 PM, John Chludzinski <john.chludzinski_at_[hidden]> wrote:
>>> # export LD_LIBRARY_PATH
>>>
>>>
>>> # mpiexec -n 1 printenv | grep PATH
>>> LD_LIBRARY_PATH=/usr/lib/openmpi/lib/
>>>
>>> PATH=/usr/lib/openmpi/bin/:/usr/lib/ccache:/usr/local/bin:/usr/bin:/bin:/usr/local/sbin:/usr/sbin:/sbin:/home/jski/.local/bin:/home/jski/bin
>>> MODULEPATH=/usr/share/Modules/modulefiles:/etc/modulefiles
>>> WINDOWPATH=1
>>>
>>> # mpiexec -n 4 ./a.out
>>> librdmacm: couldn't read ABI version.
>>> librdmacm: assuming: 4
>>> CMA: unable to get RDMA device list
>>> --------------------------------------------------------------------------
>>> [[3598,1],0]: A high-performance Open MPI point-to-point messaging module
>>> was unable to find any relevant network interfaces:
>>>
>>> Module: OpenFabrics (openib)
>>> Host: elzbieta
>>>
>>> Another transport will be used instead, although this may result in
>>> lower performance.
>>> --------------------------------------------------------------------------
>>> librdmacm: couldn't read ABI version.
>>> librdmacm: assuming: 4
>>> librdmacm: couldn't read ABI version.
>>> CMA: unable to get RDMA device list
>>> librdmacm: assuming: 4
>>> CMA: unable to get RDMA device list
>>> librdmacm: couldn't read ABI version.
>>> librdmacm: assuming: 4
>>> CMA: unable to get RDMA device list
>>> [elzbieta:4145] *** An error occurred in MPI_Scatter
>>> [elzbieta:4145] *** on communicator MPI_COMM_WORLD
>>> [elzbieta:4145] *** MPI_ERR_TYPE: invalid datatype
>>> [elzbieta:4145] *** MPI_ERRORS_ARE_FATAL: your MPI job will now abort
>>> --------------------------------------------------------------------------
>>> mpiexec has exited due to process rank 1 with PID 4145 on
>>> node elzbieta exiting improperly. There are two reasons this could occur:
>>>
>>> 1. this process did not call "init" before exiting, but others in
>>> the job did. This can cause a job to hang indefinitely while it waits
>>> for all processes to call "init". By rule, if one process calls "init",
>>> then ALL processes must call "init" prior to termination.
>>>
>>> 2. this process called "init", but exited without calling "finalize".
>>> By rule, all processes that call "init" MUST call "finalize" prior to
>>> exiting or it will be considered an "abnormal termination"
>>>
>>> This may have caused other processes in the application to be
>>> terminated by signals sent by mpiexec (as reported here).
>>> --------------------------------------------------------------------------
>>>
>>>
>>>
>>> On Sat, Sep 15, 2012 at 2:24 PM, Ralph Castain <rhc_at_[hidden]> wrote:
>>> Ah - note that there is no LD_LIBRARY_PATH in the environment. That's the problem
>>>
>>> On Sep 15, 2012, at 11:19 AM, John Chludzinski <john.chludzinski_at_[hidden]> wrote:
>>>
>>>> $ which mpiexec
>>>> /usr/lib/openmpi/bin/mpiexec
>>>>
>>>> # mpiexec -n 1 printenv | grep PATH
>>>> PATH=/usr/lib/openmpi/bin/:/usr/lib/ccache:/usr/local/bin:/usr/bin:/bin:/usr/local/sbin:/usr/sbin:/sbin:/home/jski/.local/bin:/home/jski/bin
>>>> MODULEPATH=/usr/share/Modules/modulefiles:/etc/modulefiles
>>>> WINDOWPATH=1
>>>>
>>>>
>>>>
>>>> On Sat, Sep 15, 2012 at 1:11 PM, Ralph Castain <rhc_at_[hidden]> wrote:
>>>> Couple of things worth checking:
>>>>
>>>> 1. verify that you executed the "mpiexec" you think you did - a simple "which mpiexec" should suffice
>>>>
>>>> 2. verify that your environment is correct by "mpiexec -n 1 printenv | grep PATH". Sometimes the ld_library_path doesn't carry over like you think it should
>>>>
>>>>
>>>> On Sep 15, 2012, at 10:00 AM, John Chludzinski <john.chludzinski_at_[hidden]> wrote:
>>>>
>>>>> I installed OpenMPI (I have a simple dual core AMD notebook with Fedora 16) via:
>>>>>
>>>>> # yum install openmpi
>>>>> # yum install openmpi-devel
>>>>> # mpirun --version
>>>>> mpirun (Open MPI) 1.5.4
>>>>>
>>>>> I added:
>>>>>
>>>>> $ PATH=PATH=/usr/lib/openmpi/bin/:$PATH
>>>>> $ LD_LIBRARY_PATH=/usr/lib/openmpi/lib/
>>>>>
>>>>> Then:
>>>>>
>>>>> $ mpif90 ex1.f95
>>>>> $ mpiexec -n 4 ./a.out
>>>>> ./a.out: error while loading shared libraries: libmpi_f90.so.1: cannot open shared object file: No such file or directory
>>>>> ./a.out: error while loading shared libraries: libmpi_f90.so.1: cannot open shared object file: No such file or directory
>>>>> ./a.out: error while loading shared libraries: libmpi_f90.so.1: cannot open shared object file: No such file or directory
>>>>> ./a.out: error while loading shared libraries: libmpi_f90.so.1: cannot open shared object file: No such file or directory
>>>>> --------------------------------------------------------------------------
>>>>> mpiexec noticed that the job aborted, but has no info as to the process
>>>>> that caused that situation.
>>>>> --------------------------------------------------------------------------
>>>>>
>>>>> ls -l /usr/lib/openmpi/lib/
>>>>> total 6788
>>>>> lrwxrwxrwx. 1 root root 25 Sep 15 12:25 libmca_common_sm.so -> libmca_common_sm.so.2.0.0
>>>>> lrwxrwxrwx. 1 root root 25 Sep 14 16:14 libmca_common_sm.so.2 -> libmca_common_sm.so.2.0.0
>>>>> -rwxr-xr-x. 1 root root 8492 Jan 20 2012 libmca_common_sm.so.2.0.0
>>>>> lrwxrwxrwx. 1 root root 19 Sep 15 12:25 libmpi_cxx.so -> libmpi_cxx.so.1.0.1
>>>>> lrwxrwxrwx. 1 root root 19 Sep 14 16:14 libmpi_cxx.so.1 -> libmpi_cxx.so.1.0.1
>>>>> -rwxr-xr-x. 1 root root 87604 Jan 20 2012 libmpi_cxx.so.1.0.1
>>>>> lrwxrwxrwx. 1 root root 19 Sep 15 12:25 libmpi_f77.so -> libmpi_f77.so.1.0.2
>>>>> lrwxrwxrwx. 1 root root 19 Sep 14 16:14 libmpi_f77.so.1 -> libmpi_f77.so.1.0.2
>>>>> -rwxr-xr-x. 1 root root 179912 Jan 20 2012 libmpi_f77.so.1.0.2
>>>>> lrwxrwxrwx. 1 root root 19 Sep 15 12:25 libmpi_f90.so -> libmpi_f90.so.1.1.0
>>>>> lrwxrwxrwx. 1 root root 19 Sep 14 16:14 libmpi_f90.so.1 -> libmpi_f90.so.1.1.0
>>>>> -rwxr-xr-x. 1 root root 10364 Jan 20 2012 libmpi_f90.so.1.1.0
>>>>> lrwxrwxrwx. 1 root root 15 Sep 15 12:25 libmpi.so -> libmpi.so.1.0.2
>>>>> lrwxrwxrwx. 1 root root 15 Sep 14 16:14 libmpi.so.1 -> libmpi.so.1.0.2
>>>>> -rwxr-xr-x. 1 root root 1383444 Jan 20 2012 libmpi.so.1.0.2
>>>>> lrwxrwxrwx. 1 root root 21 Sep 15 12:25 libompitrace.so -> libompitrace.so.0.0.0
>>>>> lrwxrwxrwx. 1 root root 21 Sep 14 16:14 libompitrace.so.0 -> libompitrace.so.0.0.0
>>>>> -rwxr-xr-x. 1 root root 13572 Jan 20 2012 libompitrace.so.0.0.0
>>>>> lrwxrwxrwx. 1 root root 20 Sep 15 12:25 libopen-pal.so -> libopen-pal.so.3.0.0
>>>>> lrwxrwxrwx. 1 root root 20 Sep 14 16:14 libopen-pal.so.3 -> libopen-pal.so.3.0.0
>>>>> -rwxr-xr-x. 1 root root 386324 Jan 20 2012 libopen-pal.so.3.0.0
>>>>> lrwxrwxrwx. 1 root root 20 Sep 15 12:25 libopen-rte.so -> libopen-rte.so.3.0.0
>>>>> lrwxrwxrwx. 1 root root 20 Sep 14 16:14 libopen-rte.so.3 -> libopen-rte.so.3.0.0
>>>>> -rwxr-xr-x. 1 root root 790052 Jan 20 2012 libopen-rte.so.3.0.0
>>>>> -rw-r--r--. 1 root root 301520 Jan 20 2012 libotf.a
>>>>> lrwxrwxrwx. 1 root root 15 Sep 15 12:25 libotf.so -> libotf.so.0.0.1
>>>>> lrwxrwxrwx. 1 root root 15 Sep 14 16:14 libotf.so.0 -> libotf.so.0.0.1
>>>>> -rwxr-xr-x. 1 root root 206384 Jan 20 2012 libotf.so.0.0.1
>>>>> -rw-r--r--. 1 root root 337970 Jan 20 2012 libvt.a
>>>>> -rw-r--r--. 1 root root 591070 Jan 20 2012 libvt-hyb.a
>>>>> lrwxrwxrwx. 1 root root 18 Sep 15 12:25 libvt-hyb.so -> libvt-hyb.so.0.0.0
>>>>> lrwxrwxrwx. 1 root root 18 Sep 14 16:14 libvt-hyb.so.0 -> libvt-hyb.so.0.0.0
>>>>> -rwxr-xr-x. 1 root root 428844 Jan 20 2012 libvt-hyb.so.0.0.0
>>>>> -rw-r--r--. 1 root root 541004 Jan 20 2012 libvt-mpi.a
>>>>> lrwxrwxrwx. 1 root root 18 Sep 15 12:25 libvt-mpi.so -> libvt-mpi.so.0.0.0
>>>>> lrwxrwxrwx. 1 root root 18 Sep 14 16:14 libvt-mpi.so.0 -> libvt-mpi.so.0.0.0
>>>>> -rwxr-xr-x. 1 root root 396352 Jan 20 2012 libvt-mpi.so.0.0.0
>>>>> -rw-r--r--. 1 root root 372352 Jan 20 2012 libvt-mt.a
>>>>> lrwxrwxrwx. 1 root root 17 Sep 15 12:25 libvt-mt.so -> libvt-mt.so.0.0.0
>>>>> lrwxrwxrwx. 1 root root 17 Sep 14 16:14 libvt-mt.so.0 -> libvt-mt.so.0.0.0
>>>>> -rwxr-xr-x. 1 root root 266104 Jan 20 2012 libvt-mt.so.0.0.0
>>>>> -rw-r--r--. 1 root root 60390 Jan 20 2012 libvt-pomp.a
>>>>> lrwxrwxrwx. 1 root root 14 Sep 15 12:25 libvt.so -> libvt.so.0.0.0
>>>>> lrwxrwxrwx. 1 root root 14 Sep 14 16:14 libvt.so.0 -> libvt.so.0.0.0
>>>>> -rwxr-xr-x. 1 root root 242604 Jan 20 2012 libvt.so.0.0.0
>>>>> -rwxr-xr-x. 1 root root 303591 Jan 20 2012 mpi.mod
>>>>> drwxr-xr-x. 2 root root 4096 Sep 14 16:14 openmpi
>>>>>
>>>>>
>>>>> The file (actually, a link) it claims it can't find: libmpi_f90.so.1, is clearly there. And LD_LIBRARY_PATH=/usr/lib/openmpi/lib/.
>>>>>
>>>>> What's the problem?
>>>>>
>>>>> ---John
>>>>> _______________________________________________
>>>>> 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
>>
>>
>> _______________________________________________
>> 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
|
![[Flag]](../../../../../community/mirrors/flag-us.gif){kind=small}