A Unified Framework for Evaluating Replica Placement Algorithms

The placement of data to maximize the performance and minimize the cost of a computing system is an optimization problem that has been studied extensively in several fields, including distributed databases, storage systems and, more recently, content delivery networks. However, little has been done to compare the various approaches and their applicability to different systems. This paper introduces a framework that facilitates the evaluation of replica placement algorithms. Existing algorithms are mapped to a generic set of primitives describing both problem definitions and heuristics. Based on these primitives, contributions of proposed algorithms can be crisply identified and reported. The primitives are also used to estimate the time it takes for an algorithm to reach a decision. Using actual implementations, we show that message sizes and disk access also have to be taken into account to provide a good estimate, not only the traditional computation and message complexities. We show that most proposed algorithms are prohibitively expensive for systems with more than 104 nodes and 104 objects. We also report on comparisons between algorithms with different system models, based on the goodness of the produced placement.