Wolfram Research is introducing parallel computing support for Mathematica. Now entering its beta test phase, the upcoming Parallel Computing Toolkit will add parallel programming over a network of heterogeneous machines to the long list of programming paradigms supported in Mathematica.

“I am really excited that one can now do interactive parallel symbolic, numeric, and graphic computation entirely within Mathematica,” says Roman Maeder, creator of the Parallel Computing Toolkit, author of several books on Mathematica programming, and one of the original Mathematica developers. “One of my key motivations for writing this package was to finally make serious parallel computing truly accessible to a wide range of workgroups, labs, and classrooms.”

The Parallel Computing Toolkit brings parallel computation to anybody having access to more than one computer on a network. It implements many parallel programming primitives and includes high-level commands for parallel execution of operations such as animation, plotting, and matrix manipulation. Also supported are many popular new programming approaches such as parallel Monte Carlo simulation, visualization, searching, and optimization. The implementations for all high-level commands in the Parallel Computing Toolkit are provided in Mathematica source form and serve as templates for building additional parallel programs.

The Parallel Computing Toolkit builds on Mathematica’s advanced symbolic programming language. It is written entirely in the Mathematica language and uses Mathematica’s standard MathLink protocol to communicate between any number of Mathematica kernels. The kernels can run under any supported operating system including Unix, Linux, Windows, and Macintosh. The individual machines can be single- or multiprocessor PCs and servers connected through TCP/IP.

The Parallel Computing Toolkit supports all common parallel programming paradigms: shared or distributed memory, automatic or explicit scheduling, and concurrency including synchronization, locking, and latency hiding. It also supports failure recovery. In the event of a network, hardware, or software failure, the affected computation is reassigned.