Improving the performance of checkpointing scheme with task duplication

Checkpointing is a common technique for reducing the execution time of programs under the fault assumption. With the combination of checkpointing and task duplication, not only effective fault recovery but also perfect fault detection can be achieved. The overhead of such systems comes from two aspects:comparing and saving operation at each checkpoint, and the rollbacks caused by faults. This paper improves the method presented by Zlv and Bruck by employing incremental checkpointing. The improved method can reduce the overhead of comparing and saving operation, and moreover the rollbacks caused by latent faults can be avoided. Analysis show that thatour method exhibits better performance through comparison with that of Ziv and Bruck.