ComputeResources < WS06

---++ Compute & space resources & environment

A number of sites are graciously allowing us to use some of their compute/space.

First, you should read DirectoryStructure and SyncFilesBetweenSites

---+++ JHU

   * Compute
      * 12 nodes with 2 x 2.4Ghz processors and 4G of RAM each, running Sun !GridEngine version 6.  Sungrid job submissions can be done from machines named x13 to x24. Use the option "-l ws06afsr=true" when submitting jobs.

   * Space  
      * 200G on a central NAS server accessible over NFS

   * Testing GMTK
      * to test Karen's parallel scripts,
         * unzip _/export/ws06afsr/users/lyung/gmtktest/test_parallel_scripts_v2.tgz_ to home directory
         * checkout parallel and bin (see Partha's CvsDocumentation)
         * follow Karen's README commands 

   * Setting up your environment
      * To use the common versions of GMTK, HTK and parallel scripts,
         * add the following to your .bash_profile:  =export PATH=$PATH:/export/ws06afsr/src_checked_out/bin/`uname`/:/export/ws06afsr/src_checked_out/parallel=

---+++ Edinburgh

   * Compute -- non-exclusive (but low-competition) use of:
      * Townhill !GridEngine cluster
         * 34 quad-processor-equivalent Dell PE1425 3.2GHz nodes with 4Gb RAM per node (equivalent to 136 virtual processors)
         * https://wiki.inf.ed.ac.uk/DistributedComputing/Townhill - the headnode is called townhill
         * we can also run on the three other clusters in Edinburgh
            * see https://wiki.inf.ed.ac.uk/DistributedComputing/GridEngine for their headnode names and their memory/speed specs
            * *we will have to stop using them as soon as the MT group needs this resource*
      * Condor pool
         * Currently about 200 nodes of varying specs, but expect this to vary somewhat over the summer
         * https://wiki.inf.ed.ac.uk/DistributedComputing/Condor

   * Space -- 500 GB at /group/cstr/projects/dbns/export/ws06afsr which is NFS/AFS mounted on both the townhill cluster head & compute nodes, and all the condor pool machines. 
      * Under Grid Engine: it is not possible to ssh to the townhill nodes. Initially, we will try using files directly from /group/cstr/projects/dbns/ws06afsr but if this causes problems, then we will have to start copying static data, and possibly parameter files, to the node scratch disks
      * Under Condor: always use files directly from /group/cstr/projects/dbns/export/ws06afsr and don't copy to local scratch disk.

   * Documentation on 
      * LogInToEdinburgh
      * GridEngineAtEdinburgh
      * CondorAtEdinburgh

---+++ UIUC

   * Compute 
      * Cluster with 16 compute nodes we two 2.8Ghz Xeon CPUs and 1Gb RAM each connected by a 10gb ethernet.  Jobs can be submitted via LAVA or the sungrid engine.
      * About 6 assorted machines with two or four processors and two or four gigs of RAM.

   * Space -- The total is about 3 terabytes on the cluster.  The assorted machines also have may be 200gigs of networked available disk space.

   * login instructions:
      * ssh to ifp-32.ifp.uiuc.edu.  Your login and password will be provided shortly.  
      * To keep things organized, please keep your shared files in /cworkspace/ifp-32-1/hasegawa/jhu06/.  That directory is on a networked filesystem.  A lot more space is available on the individual compute nodes.  You can access this space via /cworkspace/c1-XX where XX is 1 through 16.  You can also connect to the compute nodes directly from the ifp-32 machine by  'ssh compute-1-XX' where XX is 1 through 16, but there should be no need to do that.  
      * You can submit jobs via the [[http://gridengine.sunsource.net/howto/basic_usage.html][sun grid engine]] or via lava.  You can also see [[https://ifp-32.ifp.uiuc.edu][various statistics about the server]].  Make sure you use https and not http if you submit your password to it.
      * to see Karen's parallel test gmtk test in action run  _./emtrain_parallel.JHU_ws06 header.emtrain.test_ in /cworkspace/ifp-32-1/hasegawa/jhu06/akantor/test_emtrain_parallel .


---+++ UW

   * Compute
      * Music cluster running pmake
         * 36 CPUs, each about 1Gflop, 4 per box, 4GB per box

   * [[First log in]]
   * [[Using pmake on music]]

 
---+++++ Fine print

   * set ALLOWTOPICVIEW = %MAINWEB%.WorkshopGroup

-- Main.KarenLivescu - 12 May 2006
Find topic

WS06 topics

Members' area

Tools

Help!
