Optimal Allocation Policies for Mobile Agents

Abstract

This paper examines a distributed system where users employ a mobile software agent to perform a sequence of tasks associated with different network nodes. Each operation can be carried out either locally or remotely, and may or may not involve moving the agent from one node to another; in general, all these options have different costs. The problem is to determine the optimal agent allocation policy, for a given cost structure and pattern of user demand. The methodology adopted is that of Markov Decision Processes. Two numerical approaches are presented for the general problem, and a closed-form solution is obtained in a non-trivial special case.