Availability of Globally Distributed Nodes: An Empirical Evaluation

Abstract

Dependability models of distributed systems are often parameterised by the failure characteristics of the nodes that form a system. For realistic results, these parameters must be estimated accurately, for example, based on evaluations of real-world systems. We empirically evaluate over 400 globally distributed nodes of the PlanetLab research cluster and estimate the popular parameters mean-time-to-failure, mean-time-to-repair, availability, and failure correlation coefficients. We fit the resulting empirical distributions by simple theoretical distributions and find that the mean-time-to-failure, the availability, and the failure correlation coefficient correlate with the geographical distance between nodes.