HPC2002 Result: Hewlett-Packard Company hp server rx2600 cluster (1500MHz Itanium2)

Benchmark	Reference Time	Runtime	Ratio
361.wrf_m	86400	217	399
SPECenvM2002	399

Benchmark

Reference
Time

Runtime

Ratio

361.wrf_m

86400

217

399

SPECenvM2002

399

Hardware

Hardware Vendor:

Hewlett-Packard Company

Model Name:

hp server rx2600 cluster (1500MHz Itanium2)

CPU:

Intel Itanium 2

CPU MHz:

1500

FPU:

Integrated

CPU(s) enabled:

CPU(s) orderable:

1 to 2 per node, up to 64 nodes

Primary Cache:

L1 Inst/Data: 16 KB, associativity = 4

Secondary Cache:

L2 Unified: 256 KB, associativity = 8

L3 Cache:

L3 Unified: 6144 KB, associativity = 24

Other Cache:

None

Memory:

12GB per node (12 x 1 GB DDR 266 DIMMS)

Disk Subsystem:

1x36GB 10k RPM SCSI system disk per node

Other Hardware:

See Notes section below.

Software

Parallel:

MPI

Processes-Threads:

MPI Processes:

OpenMP Threads:

N/A

Operating System:

HPUX11i-TCOE B.11.23

Compiler:

HP C/ANSI C Compiler B.11.23
HP aC++ Compiler B.11.23
HP Fortran 90 Compiler B.11.23
HP LIBF90 PHSS_29620
HP F90 Compiler PHSS_29663
HP aC++ Compiler PHSS_29655
HP C Compiler PHSS_29656
u2comp/be/plugin library PHSS_29657
HP MPI 02.00.01.00 B6060B

File System:

vxfs (system), vxfs through NFS (benchmark files)

System State:

Multi-user

Other Software:

NetCDF 3.5.0, HP MPI v2.00.01

Notes / Tuning Information
CPU(s) enabled: 48 (two per node, 24 nodes) Other Hardware: Computation Network: AB286A PCI-X 2-port Infiniband HCA for HPC AB346A 5m copper cable PCI-X Infiniband AB353A 7m copper cable PCI-X Infiniband AB291A PCI-X 12-port InfiniBand Copper Switch Topspin 96-port IB copper switch 99-00020-01 TS170 98-00045-01 12-port leaf boards (8) 98-00047-01 power supply 98-00044-01 controller module GigaBit on-board adapter for Administration and NFS PCI GigaBit card for NFS traffic (GigE-TX adapter A6825A ) NFS file server: rp5470 (PA-RISC) NFS File Server 4 PA8700 CPUs 750 MHz. 16 GB of memory 4 internal disks 73 GB Ultra2 SCSI 20 external disks 18 GB U160 SCSI striped with LVM across 4 SCSI controllers 15 external disks 73 GB FibreChannel mirrored with LVM across 2 FC controllers which contain the NFS filesystems accessed by the benchmark. These NFS filesystems are optimized for security rather than performance. File Server Network: HP ProCurve 9308 64-port copper Gigabit Ethernet Switch Built-in Gigabit Ethernet Adapters (one per node) Peak Flags: MPI mpif90 +DD64 +noppu +Ofast +Oinfo +U77 mpicc -Ae +DD64 +Ofast -DNOUNDERSCORE -DSPEC_HPG_MPI2 OPTIMIZE = ENV_SPEC_HPG_PARALLEL=MPI CPPFLAGS = -I. -C -P EXTRA_LIBS= -minshared -L${NETCDF}/lib/hpux64 -lnetcdf NETCDF = /home/clpack/netcdf-3.5.0 FPORTABILITY= -I${NETCDF}/include Alternate Source used for Peak: 361.wrf_m: module_big_step_utilities_em.F90 Improve data locality via loop interchange. Available as SPEC HPC2002 Source: env2002-src_hp-20040303.tar.gz Kernel Paramters (/stand/system): maxdsiz 0x7b03a000 maxdsiz_64bit 0x4000000000 maxssiz 0x10000000 maxssiz_64bit 0x40000000 maxtsiz 1073741824 maxtsiz_64bit 4294967296 vps_pagesize 4 vps_ceiling 64 dbc_min_pct 3 dbc_max_pct 3 Peak User Environment: use_submit_for_speed=1 submit = /home/f90pack/clust_mpirun $command clust_mpirun: mpirun -ITAPI -f appfile appfile: -h rx17 -np 2 -e MPI_FLAGS=y -e MPI_WORKDIR=$cwd $command ... -h rx40 -np 2 -e MPI_FLAGS=y -e MPI_WORKDIR=$cwd $command LSF used to initiate batch job submissions. Appfile is generated from within the LSF run. Netcdf source obtained from http://www.unidata.ucar.edu/packages/netcdf/ Netcdf built for HPUX 64 bit mode with: #!/bin/csh setenv CC '/opt/ansic/bin/cc +DD64' setenv CPPFLAGS "-D_HPUX_SOURCE -D_FILE_OFFSET_BITS=64 -Dextname" setenv FC '/opt/fortran90/bin/f90 +DD64' setenv FFLAGS -w setenv FLIBS -lU77 setenv CXX '/opt/aCC/bin/aCC +DD64' ./configure

Notes / Tuning Information

CPU(s) enabled: 48 (two per node, 24 nodes)

Other Hardware:
  Computation Network:
    AB286A PCI-X 2-port Infiniband HCA for HPC
    AB346A 5m copper cable PCI-X Infiniband
    AB353A 7m copper cable PCI-X Infiniband
    AB291A PCI-X 12-port InfiniBand Copper Switch
    Topspin 96-port IB copper switch 99-00020-01 TS170
      98-00045-01 12-port leaf boards (8)
      98-00047-01 power supply
      98-00044-01 controller module
      
    
    GigaBit on-board adapter for Administration and NFS 
    PCI GigaBit card for NFS traffic  (GigE-TX adapter A6825A )

  NFS file server:
    rp5470 (PA-RISC) NFS File Server
      4 PA8700 CPUs 750 MHz.  16 GB of memory 
      4 internal disks 73 GB Ultra2 SCSI
      20 external disks 18 GB U160 SCSI striped with 
         LVM across 4 SCSI controllers 
      15 external disks 73 GB FibreChannel mirrored 
         with LVM across 2 FC controllers which 
         contain the NFS filesystems accessed by the benchmark. 
         These NFS filesystems are optimized for 
         security rather than performance.

  File Server Network:
    HP ProCurve 9308 64-port copper Gigabit Ethernet Switch
    Built-in Gigabit Ethernet Adapters (one per node)

 Peak Flags:  MPI
   mpif90  +DD64 +noppu +Ofast   +Oinfo   +U77
   mpicc  -Ae +DD64 +Ofast -DNOUNDERSCORE -DSPEC_HPG_MPI2  
   OPTIMIZE = 
   ENV_SPEC_HPG_PARALLEL=MPI
   CPPFLAGS = -I. -C -P                               
   EXTRA_LIBS=   -minshared   -L${NETCDF}/lib/hpux64 -lnetcdf
   NETCDF   = /home/clpack/netcdf-3.5.0
   FPORTABILITY= -I${NETCDF}/include

 Alternate Source used for Peak:
     361.wrf_m: module_big_step_utilities_em.F90 
       Improve data locality via loop interchange.
       Available as SPEC HPC2002 Source: env2002-src_hp-20040303.tar.gz

Kernel Paramters (/stand/system):
   maxdsiz         0x7b03a000
   maxdsiz_64bit   0x4000000000
   maxssiz         0x10000000
   maxssiz_64bit   0x40000000
   maxtsiz         1073741824
   maxtsiz_64bit   4294967296
   vps_pagesize    4
   vps_ceiling     64
   dbc_min_pct     3
   dbc_max_pct     3
 
 Peak User Environment:
   use_submit_for_speed=1
   submit = /home/f90pack/clust_mpirun  $command

clust_mpirun:
   mpirun -ITAPI  -f appfile
     
appfile:   
    -h  rx17  -np 2 -e MPI_FLAGS=y -e MPI_WORKDIR=$cwd  $command 
            ...
    -h  rx40  -np 2 -e MPI_FLAGS=y -e MPI_WORKDIR=$cwd  $command

 LSF used to initiate batch job submissions. 
 Appfile is generated from within the LSF run.


 Netcdf source obtained from 
    http://www.unidata.ucar.edu/packages/netcdf/
 Netcdf built for HPUX 64 bit mode with:
    #!/bin/csh
    setenv CC '/opt/ansic/bin/cc +DD64'
    setenv CPPFLAGS "-D_HPUX_SOURCE -D_FILE_OFFSET_BITS=64 -Dextname"
    setenv FC '/opt/fortran90/bin/f90 +DD64'
    setenv FFLAGS -w
    setenv FLIBS -lU77
    setenv CXX '/opt/aCC/bin/aCC +DD64'
    ./configure

First published at SPEC.org on 03-Mar-2004

Generated on Wed Mar 3 21:48:23 2004 by SPEC HPC2002 HTML formatter v1.01