Research Article
Replication in Overlay Networks: A Multi-objective Optimization Approach
@INPROCEEDINGS{10.1007/978-3-642-03354-4_39, author={Osama Al-Haj Hassan and Lakshmish Ramaswamy and John Miller and Khaled Rasheed and E. Canfield}, title={Replication in Overlay Networks: A Multi-objective Optimization Approach}, proceedings={Collaborative Computing: Networking, Applications and Worksharing. 4th International Conference, CollaborateCom 2008, Orlando, FL, USA, November 13-16, 2008, Revised Selected Papers}, proceedings_a={COLLABORATECOM}, year={2012}, month={5}, keywords={Replication Multi-Objective Optimization Evolutionary Algorithms Greedy Approach}, doi={10.1007/978-3-642-03354-4_39} }- Osama Al-Haj Hassan
Lakshmish Ramaswamy
John Miller
Khaled Rasheed
E. Canfield
Year: 2012
Replication in Overlay Networks: A Multi-objective Optimization Approach
COLLABORATECOM
Springer
DOI: 10.1007/978-3-642-03354-4_39
Abstract
Recently, overlay network-based collaborative applications such as instant messaging, content sharing, and Internet telephony are becoming increasingly popular. Many of these applications rely upon data-replication to achieve better performance, scalability, and reliability. However, replication entails various costs such as storage for holding replicas and communication overheads for ensuring replica consistency. While simple rule-of-thumb strategies are popular for managing the cost-benefit tradeoffs of replication, they cannot ensure optimal resource utilization. This paper explores a multi-objective optimization approach for replica management, which is unique in the sense that we view the various factors influencing replication decisions such as access latency, storage costs, and data availability as objectives, and not as constraints. This enables us to search for solutions that yield close to optimal values for these parameters. We propose two novel algorithms, namely multi-objective Evolutionary (MOE) algorithm and multi-objective Randomized Greedy (MORG) algorithm for deciding the number of replicas as well as their placement within the overlay. While MOE yields higher quality solutions, MORG is better in terms of computational efficiency. The paper reports a series of experiments that demonstrate the effectiveness of the proposed algorithms.
