Analysis of Optimal File Migration Policies in Distributed Computer Systems

Abstract

File migration shows promise as a means of improving data processing performance in distributed systems, but practical application of this idea will require development of effective policies that will allow a system to fully realize the potentials of file migration. Deriving of optimal policies, while computationally complex, nevertheless, is essential to provide insights about how effective migration policies for large systems should be structured. In this paper, analytic properties and performance of optimal file migration policies are investigated based on a Markov decision process model of file migration policies. Optimal migration policies are compared with optimal static policies and the sufficient conditions under which file migration provides absolute improvement or no advantage over static policies are presented. Numerical experiments and simulations were performed to analyze the impact of model assumptions and system parameters on the cost improvement generated by file migration. It is shown that optimal file migration is able to generate substantial cost improvement under certain conditions and that it is robust both with respect to the initial file allocation at an initial system design/reorganization point and to impreciseness of system environments. This analysis should provide system designers and administrators guidance toward achieving effective file migration control.