[SPAM] Re: [mvapich-discuss] [SPAM] Problem building mvapich2 1.4,
IB performance is slow/broken
Craig Tierney
Craig.Tierney at noaa.gov
Tue Nov 24 16:47:05 EST 2009
Dhabaleswar Panda wrote:
> Craig - This is very surprising. Are you carrying out inter-node or
> intra-node experiments? What platform and IB cards you are using? I am not
> sure whether the CPU mapping is playing a role here. Can you try with
> different CPU mapping and see whether you get expected performance number.
>
> Thanks,
>
> DK
>
On an intranode test, the performance is in the 10GB/s range.
The platform is Nehalem/Supermicro. We are using the on-board
QDR switches. I have messed with mappings, but the test is
only using one core per host. I will try some other things.
The same build options for 1.2p1 works.
Craig
> On Tue, 24 Nov 2009, Craig Tierney wrote:
>
>> I am trying to build the production release of mvapich2-1.4.
>> Here are the system specs:
>>
>> Centos 5.3
>> OFED-1.4.1
>> Intel compilers 11.1
>>
>> When I try and run a code built with it (OMB), the transfer
>> rates are very slow. Here is the output of the run:
>>
>> [ctierney at wfe7 ~/OMB-3.1.1]$ mpirun -np 2 ./osu_bw
>> # OSU MPI Bandwidth Test v3.1.1
>> # Size Bandwidth (MB/s)
>> 1 0.00
>> 2 0.01
>> 4 0.02
>> 8 0.03
>> 16 0.07
>> 32 0.10
>> 64 0.23
>> 128 0.39
>> 256 0.91
>> 512 1.42
>> 1024 2.26
>> 2048 2.30
>>
>> If I build and run this with mvapich2-1.2p1, I get the behavior
>> I expect.
>>
>> I have built mvapich2 with:
>>
>> ./configure CC=icc CXX="icpc" F77="ifort" FC="ifort" F90="ifort" \
>> --with-rdma=gen2 \
>> --prefix=/opt/hjet/mvapich2/1.4-intel \
>> --enable-romio=yes --with-file-system=lustre \
>> --enable-g=dbg --enable-sharedlibs=gcc --enable-debuginfo \
>> --enable-threads=multiple
>>
>> Am I missing something?
>>
>> Thanks,
>> Craig
>>
>>
>>
>> --
>> Craig Tierney (craig.tierney at noaa.gov)
>> _______________________________________________
>> mvapich-discuss mailing list
>> mvapich-discuss at cse.ohio-state.edu
>> http://mail.cse.ohio-state.edu/mailman/listinfo/mvapich-discuss
>>
>
>
--
Craig Tierney (craig.tierney at noaa.gov)
More information about the mvapich-discuss
mailing list