A realistic model for runtime incremental task scheduling in distributed memory multiprocessor systems

Abstract

Even though the area of parallel computing has existed for many decades, programming a parallel system for the execution of a single application is still a challenging problem than programming a sequential program on single processor. In this paper we investigate a partitioning and scheduling strategy for splitting a non-uniform parallel program into smaller tasks and map them into different processors at runtime. We propose a parallel program model which use explicit parallelism to identify independent section and a dynamic scheduling technique to schedule those independent sections (tasks) using message passing to different processors. Finally we show the accuracy of the scheduling by balance the loads in different processors and reduce the communication cost.