Scheduling divisible loads on heterogeneous linear networks using pipelined communications

Abstract

This work considers the divisible load distribution problem on heterogeneous linear networks. A divisible load distribution determines optimal fractions of the load and assigns them to more than one processor for minimizing the parallel execution time. Two algorithms P (pipelined communication) and M (modified method) have been proposed. The algorithm P employs the pipelined communication technique in the design. The algorithm M uses a modified method to improve the algorithm P. Closed-form expressions for the parallel processing time and speed-up are derived. In homogeneous linear networks, this work demonstrates that the pipelined communication technique outperforms the send-and-receive strategy. In heterogeneous linear networks, experiments show that the proposed algorithms are better than the send-and-receive strategy. The algorithm M is better than algorithm P when the computation-to-communication ratio is large or the number of processors is small.