Maximizing Reliability with Task Scheduling In a Computational Grid Using GA Zahid Raza, Deo Prakash Vidyarthi Maximizing Reliability with Task Scheduling In a Computational Grid Using GA Zahid Raza *, Deo Prakash Vidyarthi School of Computer and Systems Sciences Jawaharlal Nehru University New Delhi-110067 {zahidraza, dpv}@mail.jnu.ac.in doi: 10.4156/ijact.vol1.issue2.5 Abstract Grid is a service aggregation of both the information and the computational resources. Scheduling becomes a challenging job in such a complex and dynamic environment as both the application and the computational resources are heterogeneous. The problem is further complicated by the fact that these resources may fail at any point of time. Thus a scheduling strategy which schedules the job based on the failure possibility of the grid constituents becomes very important for the reliable execution of the job. Genetic algorithm has evolved as an effective search tool to solve the optimization problems consisting of large search space. It is a type of evolutionary search strategies applied on the search space and is based on the principle of “survival of the fittest”. The model addresses the important issue of scheduling to provide the most reliable environment for the job execution by scheduling the job, ensuring maximum reliability to the job execution using Genetic Algorithm. The model presents a realistic picture of the grid by scheduling the job based on the reliability of the computational resources, networking resources and application (job) along with the pre- assigned workload on the nodes in order to provide the most reliable environment to the job. Simulation study proves the effectiveness of the model by comparing the performance of the model with another similar model. Keywords Computational Grid, Scheduling, Reliability, Genetic Algorithm (GA), NP problems. 1. Introduction The objective of the Grid is to utilize all available resources that are under utilized, to the maximum possible extent. It happens quite often, in any organization, that there is huge computing power available cumulatively but only a little of it is utilized to perform the computing work. Also, sometimes a job which requires high computing energy takes substantive time in producing the results due to the non availability of the computing energy. Further, sometimes the job may need a specialized computing platform which may not be available at the node where the job is submitted. A possible solution is to integrate the computing resources available at various places and enable all authorized users, to use it in a well-defined transparent manner. The hunger for the enormous computational power by various applications (e.g. weather forecasting, molecular study, numerical data processing, particle physics etc.) has led the scientific community towards the computational power sharing and cooperation, resulting in a computational grid. Grid enables the use of networks of computers as a single, unified computing tool, clustering and coupling a variety of facilities over a wide geographic region [6-8, 17]. Scheduling a job on a grid is an NP problem. Therefore, a number of models have been proposed in the literature with a scheduling strategy with one or the other objectives. These objectives are the Job completion time (Turn around time), Fault tolerance, Reliability, Quality of service (QoS), Real time completion, Resource utilization, Security etc. The proposed study is based on scheduling the job submitted for execution on a grid with the objective to maximize the reliability of the job execution. This is done by scheduling the job over the reliable resources at any given moment of time. A genetic algorithm is an effective tool to solve NP class of problems. It works on the basis of natural selection of the fittest, among the population of individuals, according to a fitness function and then evolving towards a definite goal over the generations. In each generation the individuals undergo crossover to reproduce better off-springs. By repeating the process over generations, it gradually tends to the optimization of the desired parameter. Mutation is used to avoid local optima during the optimization process to avoid false 40