On Checkpoint Latency

Checkpointing and rollback is a technique for minimizing loss of computation in presence of failures. Two metrics can be used to characterize a checkpoint-ing scheme: (i) checkpoint overhead (increase in the execution time of the application because of a checkpoint), and (ii) checkpoint latency (duration of time required to save the checkpoint). For many checkpoint-ing methods, checkpoint latency is larger than checkpoint overhead. This paper evaluates the expression for \average overhead" of the checkpointing scheme as a function of checkpoint latency and overhead. It is shown that the \average overhead" is much more sensitive to the changes in checkpoint overhead, as compared to checkpoint latency. Also, for equi-distant checkpoints, the optimal checkpoint interval is shown to be independent of the checkpoint latency.