A new computer system named Ithaca has been installed by Virginia Tech's Advanced Research Computing facility (ARC). The system is scheduled to be open to the full user community in the near future. However, ARC is actively looking for friendly users who are interested in trying out the new facility right now!
Friendly users will be given temporary accounts on Ithaca, which will not expire until sometime after the machine is opened up to access by the full community. Friendly users will be encouraged to exercise the parallel MATLAB facility provided on Ithaca, to try to parallelize their existing MATLAB programs, and to report back on problems and issues they encounter.
If you are interested in being considered as a "friendly Ithaca user", please send mail to John Burkardt at burkardt@vt.edu.
One of the special features of Ithaca is that it can run the Parallel Computing Toolbox from the MathWorks. In the current configuration, a user job can request up to 64 cooperating copies of MATLAB at a time, and ARC plans to upgrade this to 128 in the near future.
While MATLAB is heavily used on the Virginia Tech campus, the ability to run MATLAB in parallel has so far been limited to desktop machines and small experimental clusters. As Ithaca becomes available to the entire community, many MATLAB users will finally be able to experience both the trials and benefits of parallel programming.
As an example, a simple test program for counting prime numbers was run on a single core of Ithaca, and required 8.7 minutes (517 seconds). As the number of cores was increased, the run time dropped in a fairly regular way, so that with 64 cores, the same program ran in 10 seconds, or about 50 times faster.
Athough MATLAB can be used interactively, large scale computations generally require the preparation and execution of a program that executes with little or no user input. For parallel MATLAB, this method of indirect execution is almost always necessary. However, one advantage of this fact is that it is possible for users to customize their desktop copy of MATLAB by adding an "ithaca" parallel configuration. This configuration then allows the user to submit jobs to Ithaca and to examine the results, all without ever logging in directly. Instead, MATLAB handles the details of copying data and program files up to Ithaca, requesting the necessary computational resources, and returning the output to the user. Information about how to set up this configuration is available in the document Enabling Remote Submission of MATLAB Jobs.
MathWorks documentation for the Parallel Computing Toolbox is available at "http://www.mathworks.com/products/parallel-computing/" .
In cooperation with Virginia Tech's FDI (Faculty Development Initiative), ARC will be presenting or hosting workshops, of increasing levels of detail, on parallel MATLAB. The first workshop will be a simple introduction; the later ones will be presented by the MathWorks. The tentative schedules and titles are:
If you are interested in attending the training classes, please check http://www.fdi.vt.edu , the web site for Virginia Tech's Faculty Development Institute. Check under "Spring 2010 Short Courses".
Ithaca is an IBM iDataPlex 84 node system. Each node has a pair of Intel Nehalem processors, each of which in turn has 4 cores. This means the overall system includes 672 cores. The Nehalem processors run at 2.26 GHz.
Every node has at least 24 GB of RAM; 10 of the nodes have been set up with 48 GB of RAM, to facilitate large memory jobs.
Ithaca shares the file system used by System X. The operating system is a version of Linux. Parallel programming is suppored with MPI, OpenMP and parallel MATLAB. Jobs will be submitted through the same ARC queueing system used for System X.
Compilers on Ithaca include:
Applications available on Ithaca include:
You can return to the HTML web page.