Object-Oriented Parallel Processing with Mentat

摘要

Mentat is an object-oriented parallel processing system designed to address three problems that face the parallel computing community, the difficulty of writing parallel programs, the difficulty achieving portability of those programs, and the difficulty exploiting contemporary heterogeneous environments. Writing parallel programs by hand is more difficult than writing sequential programs. The programmer must manage communication, synchronization, and scheduling of tens to thousands of independent processes. The burden of correctly managing the environment often overwhelms programmers, and requires a considerable investment of time and energy. If parallel computing is to become mainstream it must be made easier for the average programmer. Otherwise, parallel computing will remain relegated to specialized applications of high value where the human investment required to parallelize the application can be justified. A second problem is that once a code has been implemented on a particular MIMD architecture, it is often not readily portable to other platforms; the tools, techniques, and library facilities used to parallelize the application may be specific to a particular platform. Thus, considerable effort must be re-invested to port the application to a new architecture. Given the plethora of new architectures and the rapid obsolescence of existing architectures, this represents a continuing time investment. One can view the different platforms as one dimension of a two dimensional space, where the other dimension is time. One would like the implementation to be able to cover a large area in this space in order to amortize the development costs.