Task Scheduling in Distributed Systems using Discrete Particle Swarm Optimization

Finding an optimal schedule of tasks for an application in distributed environment is critical in general. Task assignment is an extremely NP complete problem. This type of problem can be resolved by heuristic algorithms efficiently because the traditional methods such as dynamic programming and the back tracking need more time for solving this NP complete problem. Particle Swarm Optimization (PSO) is a newly developed meta-heuristic global optimization technique. It was originally designed only for continuous optimization problems. In task scheduling, the particles are represented as discrete values. It is obvious that the classical PSO cannot be used to solve discrete problems directly because its positions are real-valued. Some conversion techniques are needed to operate PSO in discrete domain. This paper presents a modified PSO called Discrete PSO (DPSO). In DPSO, no conversion techniques are needed because the velocity and positions are redefined to operate the PSO in a discrete domain directly. In this paper, the scheduler aims at minimizing make span, flow time and reliability cost simultaneously in distributed systems for scheduling of independent tasks using DPSO. Benchmark instances of Expected Time to Complete (ETC) model are used to test the DPSO. Based on the simulations and comparisons, the DPSO algorithm is viable approach for the task scheduling problem. KeywordDistributed system, Heterogeneous systems, Heuristic, Task Scheduling, Particle Swarm Optimization