Matrix Generators

Many of the matrices found in the Matrix Market were obtained by running application codes which generated them. In many cases these codes are parameterized so that a wide range of matrices could actually be produced. A trivial example might be a code that generates a discrete 3D Laplacian for any set of grid sizes. Such a code could be used to generate matrices with the same general structure, but of any order. In other cases matrices with differing numerical properties could be obtained by varying various parameters, such the Reynolds number in a fluids computation. Such codes are available with the SPARSKIT distribution, for example. A service like Matrix Market could provide access to such codes in a variety of ways.

• Downloading Source Code. If freely available, the source code of a matrix generator itself could be simply made available for downloading. This would be useful for embedding the generator within a larger code for which tested sparse linear algebra software over a range of parameters presented by the generator.
• Remote Execution Service. If a user were only interested in a few particular instances of the parameter set, then having the ability run the generator remotely would be useful. In that case, an HTML form would be presented to specify the parameters of interest; submitting the form would cause the generator to run at the Matrix Market site, after which the generated matrix would be returned to the user. Such a service could also be used to provide access to proprietary matrix generators (with permission of the authors, of course).
• Executable WWW Content. A third option is to provide the matrix generator as executable WWW content such as provided by Java (Flanagan, 1996). In this case, the generator is downloaded to the user's WWW browser, where it queries the user for values of its parameters, and then executes on the user's local machine. This is a more scalable approach than providing a remote execution service, but will present some challenges in that there has yet been little experience in developing portable, reliable and efficient floating-point applications in this environment.