Performance analysis of two time-based coordinated checkpointing protocols
Gerard Patrick Kavanaugh, William H. Sanders · 2002
Time-based checkpointing protocols are a recently proposed way to improve a system's dependability. They claim to have the advantages of coordinated protocols without the normal costs of coordination. This paper investigates that claim, by analyzing and comparing two time-based checkpointing protocols. The analysis is performed by determining the forward progress of a system using each protocol, and it is described in such a way as to be easily modifiable for other time-based protocols. By carefully analyzing the behavior of each protocol between renewal points, we are able to obtain a closed-form expression for the forward progress of the two protocols considered. We also determine the checkpoint interval value that will maximize forward progress. A validation of the analytical model is then performed via a detailed simulation. The results obtained from the model show the advantages and disadvantages of each protocol.