AIST Testbed Resource Information

F32 is a 260 nodes dual Xeon 3.0GHz Linux Cluster whose
interconnection is Giga-bit Ethernet.  F32 is divided into the
following 4 partitions each of which is contributed to specific group
of users.  In each partition, the Globus gatekeeper is running on User
Service Nodes and computing nodes can be utilized via the Globus
jobmanager (jobmanager-sge).

fp1
  
  fp1 is contributed to all users.
  User Service Node:  fsvc001.asc.hpcc.jp (2 X Xeon 3.0GHz)
  Computing Nodes:    fcmp000.asc.hpcc.jp (2 X Xeon 3.0GHz)
                      :
                      fcmp063.asc.hpcc.jp (2 X Xeon 3.0GHz)
                      64 nodes / 128 cpus
  
  NOTE: Some computing nodes may be temporarily unavailable for
        administrative reason.
        The nodes of fp1 will be assigned to fp2 or fp3 if the
        nodes of fp2 or fp3 are unavailable.
        So the numbmer of nodes may be fewer and discontinuous.
  
fp2
  
  fp2 is limited to users who are involved in joint project with AIST.
  User Service Node:  fsvc002.asc.hpcc.jp (2 X Xeon 3.0GHz)
  Computing Nodes:    fcmp064.asc.hpcc.jp (2 X Xeon 3.0GHz)
                      :
                      fcmp127.asc.hpcc.jp (2 X Xeon 3.0GHz)
                      64 nodes / 128 cpus
  
fp3
  
  fp3 is limited to users who are involved in joint project with AIST.
  User Service Node:  fsvc003.asc.hpcc.jp (2 X Xeon 3.0GHz)
  Computing Nodes:    fcmp128.asc.hpcc.jp (2 X Xeon 3.0GHz)
                      :
                      fcmp255.asc.hpcc.jp (2 X Xeon 3.0GHz)
                      128 nodes / 256 cpus
  
fp4
  
  fp4 is limited to developers in AIST.
  User Service Node:  fsvc004.asc.hpcc.jp (2 X Xeon 3.0GHz)
  Computing Nodes:    fcmp256.asc.hpcc.jp (2 X Xeon 3.0GHz)
                      :
                      fcmp259.asc.hpcc.jp (2 X Xeon 3.0GHz)
                      4 nodes / 8 cpus
  
  
   (note)
   You must submit jobs to compute node with SGE(qsub) or GT(PreWS-GRAM or WS-GRAM).
   Don't execute any long running application directly on frontend node.
  

• RedHat Linux 8.0
• GCC Ver. 3.3.3
• PGI High-Performance Compilers Ver. 5.2
• Intel C++/Fortran Compiler Ver. 8.0 (Ver. 7.1 is also installed in /opt/intel/compiler70)
  • License files of Intel Compilers are already installed in F32.
  
  
• Globus Toolkit 4.0.3 / 3.2 / 2.4.3
  • GLOBUS_LOCATION = /usr/local/gt (Ver. 4.0.3)
  • GLOBUS_LOCATION = /usr/local/GT/gt-4.0.3-gcc-3.3.3 (same as above)
  • GLOBUS_LOCATION = /usr/local/GT/gt-3.2-gcc-3.3.3 (Ver. 3.3.3)
  • GLOBUS_LOCATION = /usr/local/GT/gt-2.4.3-gcc-3.3.3 (Ver. 2.4.3)
  
  
  jobmanager-sge is available on every User Service Nodes.
  Globus Toolkit on F32 is configured to use GridMPI instead of MPICH as backend for mpi jobs.
  If you submit "jobtype=mpi" job, you have to use GridMPI. If you have MPI applications which are invoked by Globus jobmanager, they must be re-compiled with GridMPI.
  We do not support MPICH as backend of the Globus Toolkit and Ninf-G. When you use WS GRAM, please submit jobs with "-factory-type SGE" option.

       $ globusrun-ws -submit -streaming -factory (hostname) -factory-type SGE -job-command (command)
       

• Globus Packaging Toolkit 3.0.1
  • GPT_LOCATION = /usr/local/gpt
• Jobmanager
  • Grid Engine 6.0u1
    • SGE_ROOT = /usr/local/sge_root
    • master node: fsvc001
    • submit node: every User Service Nodes
  
  
• Ninf-G 4.2.0 / 2.4.0 / 2.3.0
  • NG_DIR = /usr/local/ngv4 (Ver. 4.2.0)
  • NG_DIR = /usr/local/NGV4/4.2.0-gt-4.0.3-gcc-3.3.3/gcc32dbgpthr (same as above)
  • NG_DIR = /usr/local/ngv2 (Ver. 2.4.0)
  • NG_DIR = /usr/local/NGV2/2.4.0/gcc32dbgpthr (same as above)
  • NG_DIR = /usr/local/NGV2/2.3.0/{gcc32,gcc32dbg,gcc32dbgpthr,gcc32pthr} (Ver. 2.3.0)
  
  
• GridMPI
  • /usr/local/gridmpi (Ver. 0.11)
  • /usr/local/MPI/gridmpi-0.11-gcc-3.3.3 (same as above)
  • /usr/local/MPI/gridmpi-0.11-intel-8.0
  
  
  For detailed instruction of GridMPI, please refer the "Simple and Startup Usage" and "Commands" sections in http://www.gridmpi.org/gridmpi-0-11/
  If you want to use GridMPI with SGE, make and submit a job-script like below.

        1) make a job-script (ex. test.sh)
           (For example, run 8 jobs of "/path/to/your/job" by bash)
  
             #!/bin/sh
             #$ -S /bin/bash
             #$ -pe mpich 8
  
             /usr/local/gridmpi/bin/mpirun -np $NSLOTS -c $TMPDIR/machines /path/to/your/job
  
        2) submit test.sh
  
             $ qsub test.sh
       

• MPICH 1.2.6
  • ch_p4 device
    • /usr/local/mpich (Ver. 1.2.6)
    • /usr/local/MPI/mpich-1.2.6/p4-gcc-3.3.3 (same as above)
    • /usr/local/MPI/mpich-1.2.6/p4-intel-8.0
    • /usr/local/MPI/mpich-1.2.6/p4-pgi-5.2
    • /usr/local/MPI/mpich-1.2.6/p4-gcc-3.3.3-noshared (configured with --with-comm=shared option)
    • /usr/local/MPI/mpich-1.2.6//p4-intel-8.0-noshared (configured with --with-com=shared option)
  • globus2 device (built with GT 2.4.3)
    • /usr/local/mpich-g2 (Ver. 1.2.6)
    • /usr/local/MPI/mpich-1.2.6/g2-gcc32-gcc-3.3.3 (same as above)
    • /usr/local/MPI/mpich-1.2.6/g2-gcc32dbg-gcc-3.3.3
    • /usr/local/MPI/mpich-1.2.6/g2-gcc32-intel-8.0
    • /usr/local/MPI/mpich-1.2.6/g2-gcc32dbg-intel-8.0
• MPICH2
  • /usr/local/mpich2
  • /usr/local/MPI/mpich2-1.0/gcc-3.3.3 (same as above)
  • /usr/local/MPI/mpich2-1.0/intel-8.0

 a5:49:02:f5:16:a1:35:e9:49:be:d8:52:15:df:31:b7 (RSA Key)
 24:31:4f:ea:6d:70:e2:00:b9:d3:0c:7f:57:6f:58:d0 (DSA Key)

SAKURA Cluster consists of a frontend node and 16 compute nodes which have
global IP addresses.

  Frontend node:  sakura.hpcc.jp (163.220.27.104)
                  dual AMD Opteron 248 2.2GHz, 2 Gbytes Memory

  Compute node:   sakura0[0-f].hpcc.jp (163.220.27.[105-120])
		  dual AMD Opteron 244 1.8GHz, 3 Gbytes Memory

   (Note)
   Job submission for the PRAGMA routine-basis experiment is limited to the
   specific nodes (sakura0[0-d]).
   
   You must submit jobs to compute node with SGE(qsub) or GT(PreWS-GRAM or WS-GRAM).
   Don't execute any long running application directly on frontend node.

• CentOS release 5 (Final)
  
• GNU GCC (Version 4.1.2): /usr/bin/gcc
• Intel C++ compiler (Ver.9.1): /usr/local/intel/cc/9.1.047/
• Intel Fortran compiler (Ver.9.1): /usr/local/intel/fc/9.1.043/
• Globus Toolkit 4.0.4
  • GLOBUS_LOCATION = /usr/local/gt (symlink to /usr/local/GT/gt-4.0.4-gcc-4.1.1)
  • GLOBUS_LOCATION = /usr/local/GT/gt-4.0.4-gcc-4.1.1 (compiled with gcc)
  • The Jobmanager of PreWS GRAM are "jobmanager-sge" and "jobmanager-fork".
    The default is "jobmanager-sge". (you can use globus-job-run and globusrun as is)
    (ex.) $ globus-job-run (hostname) (command)
  • The Jobmanager of WS GRAM are "Fork" and "SGE". The default is "Fork".
    When you use WS GRAM, please submit jobs with "-factory-type SGE" option.
    (ex.) $ globusrun-ws -submit -streaming -factory (hostname) -factory-type SGE -job-command (command)
  • If you submit "jobtype=mpi" job, jobmanager-sge use GridMPI.
  
  
• Ninf-G 4.2.1
  • NG_DIR = /usr/local/ngv4 (symlink to /usr/local/NGV4/ng-4.2.1-gt-4.0.4-gcc-4.1.1/gcc64dbgpthr)
  • NG_DIR = /usr/local/NGV4/ng-4.2.1-gt-4.0.4-gcc-4.1.1/gcc64dbgpthr (compiled with "flavor=gcc64dbgpthr")
  
  
• GridMPI
  • /usr/local/gridmpi (symlink to /usr/local/MPI/gridmpi-2.0-gcc-4.1.2)
  • /usr/local/MPI/gridmpi-2.0-gcc-4.1.2 (compiled with gcc 4.1.2)
  For detailed instruction of GridMPI, please refer the "Simple and Startup Usage" and "Commands" sections in http://www.gridmpi.org/gridmpi-2-x/

 83:cf:92:5b:e1:ed:6e:d0:4b:3f:ff:a5:4c:07:6f:13 (RSA Key)
 57:1c:c7:6c:af:69:3a:32:0b:e0:15:c0:29:03:2c:73 (DSA Key)