Overlay routing under geographically correlated failures in distributed event-based systems

Abstract

In this paper we study the problem of enabling uninterrupted delivery of messages between endpoints, subject to spatially correlated failures in addition to independent failures. We developed a failure model-independent algorithm for computing routing paths based on failure correlations using both a-priory failure statistics together with available real-time monitoring information. The algorithm provides the most cost-efficient message routes that are potentially comprised of multiple simultaneous paths. We also designed and implemented an Internet-based overlay routing service that allows applications to construct and maintain highly resilient end-to-end paths. We have deployed our system over a set of geographically distributed Planetlab nodes. Our experimental results illustrate the feasibility and performance of our approach. © 2010 Springer-Verlag.