Benchmark for MPI/SX based on threads and processes,
and MPICH on the NEC SX-4

                          Latency      Throughput with message length
                                       0.064     1   16   256  2048 kB
------------------------------------------------------------------------
MPI/SX 8.1 (on processes) 0.01019 ms    4.95  78.3  878  2568  4660 MB/s
MPI/SX 7.2 (on processes) 0.01492 ms    3.76  54.6  599  2191  4400 MB/s
MPI/SX 7.1 on threads     0.04210 ms    1.45  22.3  347  3411  6513 MB/s
MPI/SX 7.1 on processes   0.00850 ms    6.27  95.3 1100  3314  3368 MB/s
MPI/SX 6.2 on threads     0.03689 ms    1.56  25.3  381  2853  4228 MB/s
MPI/SX 6.2 on processes   3.22009 ms    0.02   0.3    4    13    16 MB/s
MPICH on SX-4 (processes) 0.03282 ms    1.61  10.5  166   965  2144 MB/s

Benchmark protocols:
--------------------
1. MPI/SX 8.1 based on processes

SUPER-UX hwwsx4 8.1 Rev1 SX-4
Tue Mar  2 16:30:13 1999

  This Program tries to masure the bandwidth and latency of this
  SX4 system using the MPI library. The Program needs to run on two nodes.
  All timings are in milli seconds, bandwidth in MBytes per second

  Latency:
     average:   0.01019   min:   0.01018 max:   0.01021

  Bandwidth:
     msg len       average        min          max        avr.-time 1 msg
      bytes        mbyte/s      mbyte/s      mbyte/s        milli sec
           1        0.07500       0.07432       0.07505         0.01272
           2        0.14772       0.14755       0.14781         0.01291
           4        0.30491       0.30438       0.30521         0.01251
           8        0.61897       0.61713       0.62026         0.01233
          16        1.23777       1.23426       1.24053         0.01233
          32        2.47681       2.46933       2.48105         0.01232
          64        4.95156       4.93887       4.96210         0.01233
         128        9.88154       9.86581       9.90252         0.01235
         256       19.75670      19.73481      19.78022         0.01236
         512       39.41868      39.39701      39.47089         0.01239
        1024       78.33737      78.27613      78.42280         0.01247
        2048      121.70990     120.22585     121.83477         0.01605
        4096      241.03539     239.61196     242.33924         0.01621
        8192      471.46507     468.98262     474.20526         0.01657
       16384      877.97735     875.78844     878.45385         0.01780
       32768     1469.67208    1466.53944    1470.17313         0.02126
       65536     2084.71063    1932.99766    2094.27008         0.02998
      131072     1930.56434    1863.44663    1935.93151         0.06475
      262144     2567.68064    2566.73863    2568.24688         0.09736
      524288     3732.08134    3727.69348    3732.84385         0.13397
     1048576     4305.09632    4304.53371    4305.47766         0.23228
     2097152     4660.02493    4633.80434    4662.84231         0.42918
     4194304     5684.35456    5665.15212    5685.56656         0.70369


2. MPI/SX 7.2 based on processes

hwwsx4 7% /usr/local/mpi/bin/mpirun -np 2 ./latbnd
Fri Aug 22 16:58:20 1997
SUPER-UX unix 7.1 Rev1 SX-4

  This Program tries to measure the bandwidth and latency of this
  SX4 system using the MPI library. The Program needs to run on two nodes.
  All timings are in milli seconds, bandwidth in MBytes per second
  Latency:
     average:   0.01492   min:   0.01489 max:   0.01497

  Bandwidth:
     msg len       average        min          max        avr.-time 1 msg
      bytes        mbyte/s      mbyte/s      mbyte/s        milli sec
           1       0.03408      0.03400      0.03413         0.02798
           2       0.06771      0.06758      0.06784         0.02817
           4       0.23535      0.23468      0.23592         0.01621
           8       0.47072      0.46954      0.47191         0.01621
          16       0.93961      0.93743      0.94160         0.01624
          32       1.87881      1.87293      1.88389         0.01624
          64       3.75881      3.74857      3.76562         0.01624
         128       7.44438      6.43654      7.52344         0.01640
         256      14.84651     12.82718     14.98702         0.01644
         512      29.18494     29.13303     29.27112         0.01673
        1024      54.55776     54.45354     54.67345         0.01790
        2048      86.35485     85.50687     86.75209         0.02262
        4096     169.73364    168.82693    170.02818         0.02301
        8192     325.01003    323.52576    325.79957         0.02404
       16384     599.26822    597.42905    600.28737         0.02607
       32768    1035.41254   1032.50717   1037.28242         0.03018
       65536    1628.56180   1624.33295   1630.92831         0.03838
      131072    1537.46331   1527.99532   1540.47379         0.08130
      262144    2191.46317   2187.59406   2193.26976         0.11408
      524288    3145.78842   3139.71217   3149.28624         0.15894
     1048576    3881.47236   3845.10483   3889.74290         0.25763
     2097152    4399.74196   4396.30136   4403.89727         0.45457
 

3. MPI/SX 6.2 based on threads
 
hwwsx4 54% mpisx -p 2 -thread -e latbnd_7.1
Tue Oct 21 13:13:59 1997
SUPER-UX unix 7.1 Rev1 SX-4

  This Program tries to measure the bandwidth and latency of this
  SX4 system using the MPI library. The Program needs to run on two nodes.
  All timings are in milli seconds, bandwidth in MBytes per second
  Latency:
     average:   0.04210   min:   0.04073 max:   0.06506

  Bandwidth:
     msg len       average        min          max        avr.-time 1 msg
      bytes        mbyte/s      mbyte/s      mbyte/s        milli sec
           1       0.02133      0.02108      0.02188         0.04470
           2       0.04324      0.04298      0.04353         0.04411
           4       0.09067      0.09015      0.09108         0.04207
           8       0.18111      0.17904      0.18254         0.04213
          16       0.36001      0.35674      0.36291         0.04238
          32       0.72619      0.71465      0.73140         0.04202
          64       1.45664      1.44970      1.46117         0.04190
         128       2.76518      2.28909      2.92218         0.04415
         256       5.83433      5.81740      5.85190         0.04185
         512      11.60517     11.39807     11.68007         0.04207
        1024      22.33643     13.27015     23.31923         0.04372
        2048      46.58918     46.41516     46.72846         0.04192
        4096      92.86786     92.46028     93.21487         0.04206
        8192     174.38940    126.63909    184.30634         0.04480
       16384     346.56918    222.83693    368.37164         0.04508
       32768     687.15583    683.64596    689.48425         0.04548
       65536     1264.41739    1260.66016    1271.12078         0.04943
      131072     2181.08903    2055.76776    2201.84650         0.05731
      262144     3410.85151    3307.25732    3420.54054         0.07330
      524288     4639.36714    4315.25041    4682.78765         0.10777
     1048576     5574.02311    4822.24877    5825.55944         0.17940
     2097152     6513.84975    6026.08810    6627.33273         0.30704
 
 
4. MPI/SX 6.2 based on processes
 
hwwsx4 52% mpisx -p 2 -e latbnd_7.1
Tue Oct 21 13:13:34 1997
SUPER-UX unix 7.1 Rev1 SX-4

  This Program tries to measure the bandwidth and latency of this
  SX4 system using the MPI library. The Program needs to run on two nodes.
  All timings are in milli seconds, bandwidth in MBytes per second
  Latency:
     average:   0.00850   min:   0.00830 max:   0.01565

  Bandwidth:
     msg len       average        min          max        avr.-time 1 msg
      bytes        mbyte/s      mbyte/s      mbyte/s        milli sec
           1       0.02826      0.00258      0.05948         0.03375
           2       0.11675      0.11614      0.11722         0.01634
           4       0.37563      0.36398      0.38676         0.01016
           8       0.72474      0.70182      0.75697         0.01053
          16       1.56875      1.56393      1.57052         0.00973
          32       3.13733      3.12488      3.14282         0.00973
          64       6.27014      6.25429      6.28572         0.00973
         128      12.52648     12.41494     12.56850         0.00974
         256      24.91975     24.43752     25.10563         0.00980
         512      48.86719     47.28660     49.97965         0.00999
        1024      95.33615     91.92733     99.02947         0.01024
        2048     193.37424    190.05927    194.13992         0.01010
        4096     356.52418    351.62716    376.35359         0.01096
        8192     651.59484    551.22164    672.30911         0.01199
       16384     1099.67940    943.30313    1141.86682         0.01421
       32768     1733.74739    1690.91958    1753.66020         0.01802
       65536     2370.98246    2348.69374    2408.32949         0.02636
      131072     2927.66101    2917.26168    2932.92269         0.04270
      262144     3313.54105    3304.50630    3315.51507         0.07545
      524288     3548.65788    3543.01035    3562.68811         0.14090
     1048576     3138.20570    2977.76676    3150.26381         0.31865
     2097152     3367.63555    3304.65845    3377.32107         0.59389
 

5. MPI/SX 6.2 based on threads

hwwsx4 56% mpisx -p 2 -thread -e latbnd
Fri Aug  2 12:24:52 1996
SUPER-UX unix 6.1 1 SX-4

  This Program tries to measure the bandwidth and latency of this
  SX4 system using the MPI library. The Program needs to run on two nodes.
  All timings are in milli seconds, bandwidth in MBytes per second
  Latency:
     average:   0.03689   min:   0.03675 max:   0.03720

  Bandwidth:
     msg len       average        min          max        avr.-time 1 msg
      bytes        mbyte/s      mbyte/s      mbyte/s        milli sec
           1       0.02504      0.02493      0.02529         0.03809
           2       0.04976      0.04950      0.05000         0.03833
           4       0.09919      0.09888      0.09954         0.03846
           8       0.19512      0.15145      0.19889         0.03910
          16       0.39471      0.37022      0.39723         0.03866
          32       0.79086      0.78808      0.79604         0.03859
          64       1.55762      1.55241      1.57045         0.03918
         128       3.16456      3.15182      3.17595         0.03857
         256       6.33719      6.30227      6.37998         0.03853
         512      12.67049     12.62322     12.71910         0.03854
        1024      25.32882     25.19606     25.48325         0.03856
        2048      49.75779     49.60763     49.98290         0.03925
        4096      99.07479     98.77871    100.04738         0.03943
        8192     196.20188    195.80528    196.54879         0.03982
       16384     381.08052    378.50771    382.91438         0.04100
       32768     704.97065    703.83094    706.32962         0.04433
       65536    1224.04918   1220.89707   1228.96020         0.05106
      131072    1976.06527   1973.06944   1981.51013         0.06326
      262144    2852.95005   2849.59171   2856.79895         0.08763
      524288    3641.78119   3498.60755   3663.34361         0.13730
     1048576    4243.07898   4234.58470   4252.35396         0.23568
     2097152    4228.10806   4208.13668   4247.36543         0.47302
 
 
6. MPI/SX 6.2 based on processes

hwwsx4 58% mpisx -p 2 -e latbnd
Fri Aug  2 12:25:30 1996
SUPER-UX unix 6.1 1 SX-4

  This Program tries to measure the bandwidth and latency of this
  SX4 system using the MPI library. The Program needs to run on two nodes.
  All timings are in milli seconds, bandwidth in MBytes per second
  Latency:
     average:   3.22009   min:   2.84144 max:   3.61391

  Bandwidth:
     msg len       average        min          max        avr.-time 1 msg
      bytes        mbyte/s      mbyte/s      mbyte/s        milli sec
           1       0.00031      0.00029      0.00035         3.10815
           2       0.00060      0.00056      0.00066         3.17628
           4       0.00122      0.00117      0.00130         3.11742
           8       0.00243      0.00225      0.00264         3.14459
          16       0.00493      0.00450      0.00534         3.09535
          32       0.00997      0.00901      0.01056         3.06032
          64       0.02004      0.01836      0.02149         3.04611
         128       0.03092      0.01110      0.04127         3.94787
         256       0.06967      0.06487      0.07757         3.50428
         512       0.13583      0.12594      0.14706         3.59490
        1024       0.28106      0.22523      0.35778         3.47463
        2048       0.58786      0.54054      0.67535         3.32242
        4096       1.14860      1.11946      1.23336         3.40087
        8192       2.22997      2.16282      2.34332         3.50342
       16384       4.16656      4.04280      4.31104         3.75009
       32768       7.80768      7.42769      8.19544         4.00247
       65536       8.77331      8.40643      9.03560         7.12388
      131072      10.79435     10.54361     11.21163         11.58013
      262144      13.36839     12.99053     13.90881         18.70084
      524288      15.58335     15.25815     16.07687         32.08552
     1048576      14.97687     14.02077     15.74776         66.76964
     2097152      16.18238     15.66809     16.77070         123.59122


7. MPICH on NEC SX-4, based on processes

hwwsx4 39% /usr/local/mpich/bin/mpirun -np 2 latbnd_ch
Wed Aug  7 12:57:05 1996
SUPER-UX unix 6.1 1 SX-4

  This Program tries to measure the bandwidth and latency of this
  SX4 system using the MPI library. The Program needs to run on two nodes.
  All timings are in milli seconds, bandwidth in MBytes per second
  Latency:
     average:   0.03282   min:   0.03248 max:   0.04095

  Bandwidth:
     msg len       average        min          max        avr.-time 1 msg
      bytes        mbyte/s      mbyte/s      mbyte/s        milli sec
           1       0.02803      0.02798      0.02806         0.03403
           2       0.05487      0.04151      0.05593         0.03476
           4       0.10981      0.09170      0.11116         0.03474
           8       0.22080      0.18407      0.22662         0.03455
          16       0.42955      0.36302      0.43933         0.03552
          32       0.84732      0.71081      0.85762         0.03602
          64       1.60689      1.36594      1.62617         0.03798
         128       3.48259      2.91696      3.52637         0.03505
         256       6.99006      5.85254      7.09019         0.03493
         512      13.95630     11.66785     14.11927         0.03499
        1024      10.53303     10.38540     10.55375         0.09271
        2048      21.09037     20.95989     21.94923         0.09261
        4096      41.95727     39.67791     46.11855         0.09310
        8192      82.25768     79.86838     83.08141         0.09498
       16384     166.53791    160.86073    171.73629         0.09382
       32768     316.16723    281.99656    319.96745         0.09884
       65536     521.40809    438.39233    561.38628         0.11987
      131072     615.05620    592.32100    622.45422         0.20323
      262144     964.89433    937.79855    973.65337         0.25910
      524288    1468.33206   1427.41084   1474.34979         0.34052
     1048576    1795.12006   1770.32743   1796.78880         0.55707
     2097152    2144.35689   2117.11463   2148.84057         0.93268