Failure Recovery: A Model and Experiments

Abstract

This paper presents a model of failure recovery from which we have designed and tested sets of failure recovery methods in the Phoenix system. We derive the model, document its assumptions, and then test the validity of the assumptions and predictions of the model. We present three experiments. One derives baselines for failure recovery in the Phoenix environment. The second compares the performance of two strategies for selecting failure recovery methods. The third compares the performance of an initial set of failure recovery methods with a redesigned set that is predicted to have lower expected cost.