Machine learning in a policy driven grid environment

Abstract

Policy driven Grid Computing environment is widely used in professional organisation and educational institutions when complex tasks need to be performed. Policies are designed so that the performance of the system is optimal and throughput of the Grid setup is maximum and also cater to the needs of all the users who are consumers of the Grid environment. Memory is one of the key resourcing parameter and policies involving the memory usage of a job are designed so that memory starvation is prevented. Machine Learning algorithms which finds its applications in diverse fields are used to predict the memory requirements of a job which in turn leads to better utilization of the Grid infrastructure. Two different datasets are considered in this paper from diverse fields and the memory requirements were predicted. Promising results were obtained from the proposed algorithm and this can be used in any field for identifying the memory requirement of jobs at a project/user level. © 2014 Springer-Verlag.