Dolphin Wulfkit

[Patrick Geoffray]

Joachim,

Joachim Worringen wrote:
 > Regarding the placement of the process (or: assignment of ranks) on
 > SMP-nodes is part of the peformance strategiy that an MPI library is
 > free to follow. It seems that ScaMPI does a straight round-robin mapping
 > of the process-to-node-ranks, while MPICH-GM "groups" processes on the
 > same node. Both approaches are, of course, valid.

No, an MPI library does not take this type decision, the user or the
batch queue system does. MPICH-GM is using a conf file provided by the
user, and will never take the freedom to remap the process allocation. 
In this case, the conf file used was listing each node twice grouping 
them instead of round-robin. Information like that should be part of a 
benchmarking protocol if you want to do it seriously.

 > The attachment of you results wasn't readable for me, can you post it

it was gziped ASCI. I have put them on http://www.cs.utk.edu/~pgeoffra/

 > again or send it via Mail? I will be happy adding it to the page. If
 > you have "real" results for the P4-i860 platform, please send them,
 > too, and I'll replace the other ones. As it says in the
 > disclaimer... ;-)

I have only 2 P4-i860, so I won't be able to give more than PingPong, 
but it's not at all the reported results (Supermicro P4DC6, Lanai9 200 
MHz, GM-1.5.1 with i860 optimisation enabled).

#---------------------------------------------------
# Benchmarking PingPong
# ( #processes = 2 )
#---------------------------------------------------
        #bytes #repetitions      t[usec]   Mbytes/sec
             0         1000         8.77         0.00
             1         1000         8.98         0.11
             2         1000         8.98         0.21
             4         1000         8.97         0.43
             8         1000         9.02         0.85
            16         1000         9.13         1.67
            32         1000         9.51         3.21
            64         1000         9.72         6.28
           128         1000        12.80         9.54
           256         1000        15.28        15.97
           512         1000        18.69        26.13
          1024         1000        25.27        38.64
          2048         1000        38.87        50.25
          4096         1000        59.81        65.31
          8192         1000        98.57        79.26
         16384         1000       136.94       114.10
         32768         1000       218.81       142.82
         65536          640       382.51       163.39
        131072          320       710.12       176.03
        262144          160      1364.91       183.16
        524288           80      2675.10       186.91
       1048576           40      5299.45       188.70
       2097152           20     10425.82       191.83
       4194304           10     20689.25       193.34
#=====================================================

Your disclaimer makes perfect sense, but when people reference your 
page, they "forget" to reference also the disclaimer :-)

Patrick

----------------------------------------------------------
|   Patrick Geoffray, Ph.D.      patrick at myri.com
|   Myricom, Inc.                http://www.myri.com
|   Cell:  865-389-8852          685 Emory Valley Rd (B)
|   Phone: 865-425-0978          Oak Ridge, TN 37830
----------------------------------------------------------