A general software development environment for parallel tasks scheduling and performance assessment

An interesting and cardinal software problem of parallel processing systems is task partitioning and scheduling. It is how a problem can be decomposed into parallel execution parts which are then dispatched on processors in an efficient way to minimize program processing time and communication delays. The assortment of different applications makes the task scheduling problem very stimulating and hard to solve in parallel processing systems. Thus, the efficient partitioning and scheduling of parallel or sequential programs in parallel processing systems to utilize the available processing powers has become an extremely crucial research issue in the field of parallel computing. This dissertation presents three novel heuristic scheduling methods, namely DLP, LIBRA/DLP, and LCTD, within the scope of the PARSA (PARallel program Scheduling and Assessment) parallel software development tool that is designed to address the efficient partitioning and scheduling of parallel programs on multiprocessor systems. The proposed scheduling methods are integrated to the PARSA as the Scheduling and Partitioning tool. The PARSA environment consists of a user-friendly (visual) and interactive compile-time environment for partitioning, scheduling, and performance evaluation/tuning of parallel programs on different parallel computer architectures.