Applicaion Test Procedure Using NAS Parallel Benchmarks

After your SCore cluster has started to work stable, let's try to measure a application performance of the SCore cluster using NAS Parallel Benchmarks. The NAS parallel benchmarks were developed by NASA in order to measure thier application performance of computational fluid dynamics for aerospace research. This SCore distribution includes binaries of NAS parallel benchmark (Version 2.4). This binary distribution is permited by NAS parallel benchmark division in NAS.
  1. Executable Binary Archive Location and Descriptions

    The archive of NAS parallel benchmark executable archive is /mnt/cdrom/NPB/NPB.binary.tar.bz2 when SCore distribution CD-ROM is mounted on /mnt/cdrom.

    This SCore distribution CD-ROM includes bt, cg, is, lu, mg, sp benchmark binaries except ft in NAS parallel benchmarks. The executable binary name is different by applications, classes and number of processors. The classes show sizes of benchmark, S, W, A, B, C and D (sequences from smaller size). The provided binaries are class A and B. The compliler is GCC(g77) .

    Samples of executables
    -rwxr-xr-x root/root   2062063 2003-02-04 15:14:05 bt.A.4
    -rwxr-xr-x root/root   1985116 2003-02-04 14:57:17 cg.A.4
    -rwxr-xr-x root/root   1959537 2003-02-04 16:19:37 ep.B.4
    -rwxr-xr-x root/root   1913705 2003-02-04 14:57:28 is.A.4
    -rwxr-xr-x root/root   2043303 2003-02-04 16:25:33 lu.A.4
    -rwxr-xr-x root/root   2006908 2003-02-04 16:20:56 mg.B.8
    -rwxr-xr-x root/root   2068424 2003-02-04 17:00:04 sp.B.36

    Binary descriptions "Application"."Class"."CPUs"
    Ex:bt.A.4 means bt benchmark, Class A for 4 processors.

  2. How to run

    For lu.A.4
    % scrun -nodes=4 lu.A.4

    For sp.A.4
    % scrun -nodes=4 sp.A.4

  3. How to Check Benchmark Results

    When a NAS parallel benchmark binary is executed, the benchmark program outputs exection results (ex, execution time, performance, and verification).
    Ex: is.B.8
    % scrun -nodes=8x1 is.B.8
    SCore-D 5.8,2 connected.
     NAS Parallel Benchmarks 2.4 -- IS Benchmark
     Size:  33554432  (class B)
     Iterations:   10
     Number of processes:     8
     IS Benchmark Completed
     Class           =                        B
     Size            =                 33554432
     Iterations      =                       10
     Time in seconds =                     8.00
     Total processes =                        8
     Compiled procs  =                        8
     Mop/s total     =                    41.95
     Mop/s/process   =                     5.24
     Operation type  =              keys ranked
     Verification    =               SUCCESSFUL
     Version         =                      2.4
     Compile date    =              04 Feb 2003
     Compile options:
        MPICC        = mpicc 
        CLINK        = mpicc 
        CMPI_LIB     = (none)
        CMPI_INC     = (none)
        CFLAGS       = -O4
        CLINKFLAGS   = (none)
     Please send the results of this run to:
     NPB Development Team
     If email is not available, send this to:
     MS T27A-1
     NASA Ames Research Center
     Moffett Field, CA  94035-1000
     Fax: 650-604-3957

    The output lines of Mop/s total(/s/process) and Verification are most important for evaluation. Let's try to changing number of processors, and evaluate the scalability of the program and verification.

  4. Let's Compare the Performance Results

    Investigation and evaluation section in PC cluster consourtium plans to collect and compare of benchmark results of NAS parallel benchmarks. The detail will show PC Cluster Consortium Web page and SCore-Users ML.

Trouble Shooting

Benchmark results are not good

Benchmark result is failed

$Id: npb-test.html,v 1.6 2004/11/25 05:01:07 kameyama Exp $