Load Balance Aware Genetic Algorithm for Task Scheduling in Cloud Computing

Abstract

This paper proposes to solve the task scheduling problem in cloud computing by using a load balance aware genetic algorithm (LAGA) with Min-min and Max-min methods. Task scheduling problems are of great importance in cloud computing, and become especially challenging when taking load balance into account. Our proposed LAGA algorithm has several advantages when solving this kind of problems. Firstly, by introducing the time load balance (TLB) model to help establish the fitness function with makespan, the algorithm benefits from the ability to find the solution that performs best on load balance among a set of solutions with the same makespan. More importantly, the interaction between makespan and TLB helps the algorithm to minimize makespan in the same time. Secondly, Min-min and Max-min methods are used to produce promising individuals at the beginning of evolution, leading to noticeable improvement of evolution efficiency. We evaluated LAGA on several task scheduling problems and compared with a Min-min, Max-min improved version of genetic algorithm (MMGA), which does not use the TLB strategy. The results show that LAGA can obtain very competitive results with good load balancing properties, and outperform MMGA in both makespan and TLB objectives.