High Performance Computing in Integrated Environments

Document Type


Publication Date



Computer Sciences | Physical Sciences and Mathematics


Michael Heroux, Computer Science


Iterative numerical solvers are essential in many areas of engineering. Most high performance solvers rely on lower-level programming languages for the backbone of the computation. By using newer extensions to programs like Matlab, engineers can save time and energy that would be lost to rewriting code and create more readable code that is also easier to debug. Two such extensions are examined: Star-P and the Distributed Computing Toolbox. We found that while Star-P is very easy to program, there are some applications that Star-P cannot run well. The alternative, DCT, required some knowledge of data handling, but showed better performance for each processor used. The important result is that there are always compromises made when using higher-level languages for high performance computing.