Design and performance analysis of a message scheduling scheme for WLAN-based cluster computing

Abstract

This paper proposes and analyzes the performance of a task scheduling scheme that couples network schedule for master-slave style parallel computing cluster built on top of wireless local area networks. When a transmission fails, the proposed scheme selects another data and destination that can replace the subtask scheduled on the unreachable node with minimal cost, rather than hopelessly retransmits on the bad channel. Simulation results performed via ns-2 event scheduler, show that the proposed scheme minimizes the task migration, improves the computation time by maximally 35.4 %, increases the probability of successful retransmission, and finally survives the network failure as long as at least one node is reachable at each instance of time. © Springer-Verlag Berlin Heidelberg 2006.