Allocating Services to Applications using Markov Decision Processes

Abstract

In this paper, we outline the problem of allocating instances of services to different applications. We study a simple form of this problem with three types of services and two classes of applications. We address the problem when all instances of services have the same QoS parameters and when they have different QoS parameters. The objective of our problem is to find an optimal policy for accepting requests for each class of application, to maximize the utilization of available instances of services. To do so, we used Markov decision processes and we computed the optimal policy for our problem. Also, we compared the performance of this optimal policy with complete sharing and complete partitioning methods. The results show that using Markov decision processes we can obtain an optimal policy for allocating services to applications which outperforms the other two methods.