| This thesis presents the design and implementation of a high-level parallel programming system. The system is designed to support parallel scientific computing and to unburden the user from the responsibility of discovering and managing parallelism such as partitioning, mapping, scheduling, and load balancing. Thus, it will allow a user to concentrate on solving the problem.;From the language perspective, we integrate the reuse of legacy Fortran code with the intensional programming paradigm. At runtime, coarse-grain parallelism can be exploited to allow concurrent execution of Fortran subprograms on sequential machines connected by a local or wide area network. From the implementation perspective, an application program written in this system is mapped onto a particular object-oriented parallel generator/executer abstract program architecture. At runtime, this architecture will form the object-oriented eduction engine to comply with the intensional semantic. We choose client-server and multithread computational models to achieve both pseudoparallel and true parallel computations in the distributed environment.;By making the new programming system architecture-neutral, users can avoid additional operational details, such as the additional effort required to maintain code on different platforms when implementing the program on a heterogeneous distributed environment. In such an environment, each machine operates differently and therfore requires platform-dependent implementations. Also, the program must be compiled on each machine before it can be executed as additional time and effort in program development. In our system environment the user will be able to run his/her parallel programs on a variety of hardware architectures, operating systems and user interfaces without program recompilation and manual parallel tasks distribution. |
