Optimization of file migration policies in distributed computer systems

Abstract

Dynamic file migration has been implemented in recent designs of distributed systems to support data processing activities. Use of file migration promises to improve data processing performance in distributed systems more effectively than periodic file reallocations, but actual realization of its potential benefits reqires a practical yet effective control policy. This paper examines the cost tradeoffs between dynamic file migration and static policies and presents an optimization model for selecting optimal migration policies. Inherited from the recursive relationship among successive decisions regarding the implementation of file migration operations, the optimization model has the form of a Markov decision model. The proposed model not only can be adopted by small-sized systems but also provides a foundation for effective and simple heuristics for adaptive file migration in large systems. Numeric comparisons of system performance of optimal file migration and static policies have been performed. The results show that optimal implementations of file migration operations generate performance improvements of varying degree over static designs.