[mvapich-discuss] Multi bandwidth message rate benchmark results (osu_mbw_mr)

Jonathan Perkins perkinjo at cse.ohio-state.edu
Thu Apr 7 08:33:01 EDT 2011


Johnny:
These results are surprising.  We'll take a look at the benchmark to
see what may cause this.

On Thu, Apr 7, 2011 at 4:50 AM, Johnny Devaprasad
<johnnydevaprasad at gmail.com> wrote:
> Hi all,
> My run of osu_mbw_mr provides the following results.
> $ mpirun_rsh -np 5280 -machinefile /tmp/382.1.big/machines
>  /home/cmsupport/mvapich2/osu_benchmarks/osu_mbw_mr
> # OSU MPI Multiple Bandwidth / Message Rate Test v3.3
> # [ pairs: 2640 ] [ window size: 64 ]
> # Size                    MB/s          Messages/s
> 1                        41.34         41339222.14
> 2                        78.96         39479870.86
> 4                       156.71         39176501.04
> 8                       323.37         40421330.96
> 16                      613.74         38358861.20
> 32                     1269.45         39670442.09
> 64                     2388.31         37317373.14
> 128                    4254.61         33239163.57
> 256                    7978.03         31164175.18
> 512                   15421.78         30120662.24
> 1024                  29769.91         29072181.60
> 2048                  47861.16         23369709.15
> 4096                  65596.75         16014831.90
> 8192                  70488.55          8604559.21
> 16384                 70629.90          4310907.20
> 32768                 71233.64          2173878.15
> 65536                 71501.51          1091026.41
> 131072                71659.65           546719.78
> 262144                71741.03           273670.31
> 524288                71765.61           136882.04
> 1048576              -39592.43           -37758.29
> 2097152               16099.89             7677.02
> 4194304              -11749.33            -2801.26
> Is there an explanation on why the data rate has negative values for some of
> the larger message sizes?
> Mvapich2 version: 1.6
> Number of nodes: 110
> Number of cores : 5280 (48 cores per node)
> IB Information:
> --------------------
> InfiniBand: Mellanox Technologies MT26428 [ConnectX VPI PCIe 2.0 5GT/s - IB
> QDR / 10GigE] (rev b0)
> $ ibstat
> CA 'mlx4_0'
> CA type: MT26428
> Number of ports: 1
> Firmware version: 2.7.626
> Hardware version: b0
> Regards,
> Johnny
>
> _______________________________________________
> mvapich-discuss mailing list
> mvapich-discuss at cse.ohio-state.edu
> http://mail.cse.ohio-state.edu/mailman/listinfo/mvapich-discuss
>
>



-- 
Jonathan Perkins
http://www.cse.ohio-state.edu/~perkinjo



More information about the mvapich-discuss mailing list