Skip to content

Installing and Running Star-P on the OnDemand Cluster

ATTENTION: OnDemand is in pre-production. View or subscribe to SDSC News for updates. To request access to this resource, contact consult@sdsc.edu.

Star-P is a software platform to link between MATLAB® tools and high performance parallel computing resources. The software is provided by Interactive Supercomputing Corporation. See their Web site for more information.

Star-P software is a client-server, parallel-computing platform. The server is currently installed on the OnDemand cluster. The client is installed on the user's desktop system with the MATLAB installation. Users need to have a UNIX account on the server and must have a MATLAB license for their personal workstation.

Star-P Installation

To set up the account and software environment after obtaining an allocation on the OnDemand Cluster:

  1. Update your MATLAB software to the most current version (Release 14 service pack 2 or higher).
  2. Download and install the Star-P client package from /share/apps/starp/client_software in the server system: rocks-131.sdsc.edu.
  3. Install PuTTY package if your client is installed for the Windows environment and do the following putty setup:
    1. Make sure that PuTTY is in your windows path and that you can invoke the commands plink and PuTTY from the command line.
    2. Copy the public and private key from rocks-131.sdsc.edu to your windows system
    3. Run PuTTYgen, import the private key that you copied from rocks-131
    4. Save the private key. This will create starp_rocks.ppk file. (I like that Star-P Rocks!)
    5. Start Pageant
    6. Add the key that you created to Pageant.
  4. Modify the starpd.properties in c:\Program Files\starp\config for the Windows system
  5. # matlab must be in your Windows executable search path:
    starpd.client.matlab.executable=matlab
    
    # Set this property to true to make matlab
    # communicate with starpserver over
    # an ssh tunnel.  This will probably
    # increase latency, but may be required
    # if you can only connect to the HPC via ssh.
    # Set to "true" or
    # "tunnel" to use SSH tunnelling, or
    # set to "netcat" if your ssh
    # server doesn't all port forwarding, to use a different method.
    
    starpd.client.matlab.usessh=true
    #starpd.client.matlab.usessh=false
    
    # the web address of the Admin Server (if used)
    #starpd.mcm.url=http://:/rpc/master_control_module/api
    # These parameters can be set to automate login to the admin server
    #starpd.mcm.login=joe
    #starpd.mcm.password=password
    # can be openssh, jsch, or putty
    starpd.ssh.client=putty
    
    # Set to true to use a custom version of exec to execute subprocesses,
    # which doesn't close inherited file descriptors.
    starpd.usecustomexec=false
    
    # The workgroup manager command, relative to the StarP installation directory
    #starpd.wgm.command=bin/xmlrpc_starpsession_handler.py
    starpd.wgm.command=starp.proxy
    
    # The starpd.wgm.* options specify information that would normally come from
    # the Admin Server. These can also all be specified from the commandline.
    
    # Number of processors to request. Corresponds to --numprocs option
    starpd.wgm.numprocessors=8
    
    # HPC to connect to. Corresponds to --hpcaddress option
    starpd.wgm.address=rocks-131
    
    # login name used to ssh to the HPC. Corresponds to --hpcuser
    starpd.wgm.login=dchoi
    
    # path where starp is installed on the HPC. Corresponds to --starppath
    #starpd.wgm.starppath=/usr/local/starp/2.4.0
    starpd.wgm.starppath=/home/dchoi
    
    #starpd.ssh.tunnelcmdline={0} -l {1} -L {2}:kong.isc-dev.com:{4} -p {6} isc-dev "{7}"
    starpd.ssh.tunnelcmdline={0} 
         -ssh -batch -l {1} -L {2}:compute-0-31:{4} -P {6} rocks-131 "{7}"
    
    # path where starpserver will store data.  Corresponds to --datadir
    
    starpd.wgm.data_path=.
    

Running Star-P

  1. Grab the interactive nodes using "qrsh" command:
    qrsh -pe mpi 8

    Here we are trying to grab 8 cores (= 2 OnDemand nodes) for the Star-P nodes.

  2. Excute the node parse program and follow the direction from the program.
    /share/apps/starp/starp_nodes
  3. Execute the StarP client from your desktop.

Did You Get
What You
Wanted?
Yes No
Comments