If (top != self) { window.location = 'about:blank'; }
NASA Logo, National Aeronautics and Space Administration
High-End Computing Program

+ Home > News & Publications > Feature Stories >

FEATURE STORY

NCCS Offering Parallel MATLAB on Discover Cluster 07.25.2008

By Jarrett Cohen

The NASA Center for Computational Sciences (NCCS) is offering a parallel version of MATLAB to its user community for evaluation. The software runs on the Discover computing cluster and is accessed through a special queue.

MATLAB is an interactive programming environment for developing algorithms, analyzing and visualizing data, and managing projects. Perhaps the preeminent software tool of its kind, more than one million scientists and engineers in 175 countries are currently using MATLAB.

"With the high popularity of MATLAB at Goddard Space Flight Center and other NASA centers, we want to allow our users to test and evaluate MATLAB on a high-end computing system," said Mike Rouch, an NCCS system architect with Computer Sciences Corp. (CSC). "This is not yet a production service but an evaluation to determine if we want to make it a production service."

Image of MATLAB's graphical user interface with several panes

Using MATLAB's graphical user interface, scientists can develop algorithms, analyze and visualize data, and manage projects. A parallel version of MATLAB is being offered for evaluation on the NCCS Discover cluster. Figure courtesy of The MathWorks, Inc. (Click on the image for a larger version.)

The NCCS version of MATLAB combines three products: MATLAB 7, Parallel Computing Toolbox 3.3, and MATLAB Distributed Computing Server. They are installed on Discover's Analysis and Visualization Nodes. NCCS staff chose these nodes because they have twice as much memory per processor core—2 gigabytes—as the computing portion of Discover. The 16 nodes each consist of four cores (housed in 2.6 GHz dual-core AMD Opteron chips) and 8 gigabytes of memory. The trial license can accommodate one user at a time, and the software is configured to allow using up to two nodes (16 processor cores). If the NCCS decides to put parallel MATLAB into production, a full license would allow multiple, simultaneous users.

Users with Discover accounts will have immediate access to parallel MATLAB by submitting to the Visual Queue (visit the first link below this article for instructions). NASA-funded scientists who do not yet have NCCS accounts may apply for a MATLAB user account with NCCS User Services. In coming months, the NCCS might also install parallel MATLAB on the Data Portal, which is accessible to those without computing accounts.

Most Goddard users have been running MATLAB on their individual workstations. Rouch said that good candidates for the parallel version would answer "yes" to at least one of three questions:

  • Do you have data sets that are too large for one processor?
  • Do you have tasks that can or need to be parallelized?
  • Are your problems taking so much time to run that you need to substantially speed them up?

A major advantage of parallel MATLAB for large data sets is its ability to divide a mathematical array across multiple processors. For task-parallel applications, MATLAB is especially adept at performing repetitive functions such as Monte Carlo (probability) simulations and can take advantage of parallel for-loops.

Rouch said that the transition for experienced MATLAB users should be relatively painless. "You can execute a program on parallel MATLAB without radically changing your code," he said. "MATLAB automates many of these functions."

To assist users, MATLAB vendor The MathWorks, Inc. scheduled a "Parallel Computing in MATLAB Master Class" at Goddard on August 1. It was part of a two-day "MATLAB and Simulink Technical Workshop." If there is interest, similar events may be held in the future.



For information on using parallel MATLAB and other visualization and analytics software at the NCCS, visit:
https://www.nccs.nasa.gov/user_info/primer/software/viz_and_analytics

To contact NCCS User Services, email or call:
support@nccs.nasa.gov or 301-286-9120

To apply for an NCCS account online, visit:
https://www.nccs.nasa.gov/user_info/new_user

For more information about MATLAB and related products, visit:
http://www.mathworks.com/products/matlab/index.html?ref=pfo

For questions about the workshop or MATLAB, contact:
Alison Gallagher, alison.gallagher@mathworks.com, 508-647-4626

USA.gov NASA Logo - nasa.gov