Reducing the Effects of Time Cheating on the Performance of Divisible Load Scheduling Using Analytical Hierarchy Process

Abstract

The divisible load theory (DLT) is a paradigm in the area of distributed and parallel computing. As a matter of fact, the computations and communications can be divided into some independent part in which each part can be executed separately by a processor. The problem that the processors may cheat the algorithm has examined the divisible load theory. However, the computation rate cheating issue may appear if the processors accomplish their fraction of loads with various rates. According to the literature, if the processors do not report their true computation rates, they can not obtain optimal performance. This paper focuses on this problem. This paper proposes an AHP-based divisible load scheduling method aiming to decrease the impacts of cheating on the efficiency of divisible load scheduling. The experimental results indicate the proposed method considerably reduce the impacts of cheating on the startup time, speedup, and makespan specially when a huge number of processors cheat the algorithm.