Checkpoint placement algorithms for mobile agent system

Abstract

Checkpointing is a fault tolerance technique widely used in various types of computer systems. In checkpointing, an important issue is how to achieve a good trade-off between the recovery cost and the system performance. Excessive checkpointing would result in the performance degradation due to the high costly I/O operations during checkpointing. Equidistant and equicost are two well-known checkpointing strategies for addressing this issue. However, there is no study on these strategies catering for a mobile agent (MA) system, which has different characteristics with conventional systems. In this paper, based on an analysis of the behaviours of an MA system, we find that it can be modelled as a homogeneous discrete-parameter Markov chain, which is different from the models used in conventional systems. Therefore, the analytic methods and corresponding results for conventional systems cannot be adopted directly for an MA system. Based on our proposed model, we study the equidistant and equicost checkpointing strategies and propose checkpoint placement algorithms for MA systems. Through simulations we evaluate the performance of our proposed algorithms and the result shows that the equicost strategy based algorithm is most suitable for an MA system.