Parallel programming environment and interconnection network emulation

The need to solve increasingly complex problems has put parallel programming in the forefront of Computer Science. A large body of parallel algorithms and the growing number of commercial parallel computers are the foundation of this increasingly important field. However the availability of parallel software has not flourished as much as one could hope. It is believed that its progress is hindered by the lack of cross-platform parallel programming environment. The difference between network topology used in a parallel algorithm and the one adopted by a parallel computer is one of the major obstacles in adapting the algorithm for various platforms.; This dissertation investigates the practical feasibility of network emulation algorithms and develops a realistic parallel programming environment to support network emulations. The environment is built around a new parallel programming language, iC, which supports the SIMD system and user-defined network topology. In principle, the iC language acts as a bridge between parallel programming languages based on different network topology. The iC language is furnished with a number of inter-compilers and outer-compilers. Each inter-compiler converts an iC program based on one network topology to an equivalent iC program based on another network topology; each outer-compiler translates between iC programs and equivalent parallel programs for an existing parallel computer with the similar network topology. A working prototype of the iC compilers for three different platforms has been implemented. They are nCUBE, an MIMD system with a hypercubic network, MasPar, a SIMD system with a two-dimensional array network, and SUN Sparc, a machine with a single main processor.; The research demonstrates that building a machine independent parallel programming environment is feasible. It provides a powerful experimental tool to study the applicability of parallel algorithms and emulation algorithms on different network topologies, most of which are proposed to be based on pure theoretical models. The programming environment is suitable for teaching and studying parallel programming, in particular parallel programming involving distributed memory systems. The iC compilers facilitate the study on the effect of different mapping algorithms to the program execution. Furthermore, the compilers can also be used as prototypes to develop optimizing compilers for parallel languages.