Performance analysis of checkpointing strategies

The general model unifies a variety of models that have previously been investigated and introduces a novel checkpointing strategy, the equicost strategy, which is failure-dependent and reprocessing-independent, and compares the maximum system availability resulting from the equidistant and the Equicost checkpointing strategies under Weibull distributions.

Abstract

A widely used error recovery technique in database systems is the rollback and recovery technique. This technique saves periodically the state of the system and records all activities on a reliable log tape. The operation of saving the system state is called checkpointing. The elapsed time between two consecutive checkpointing operations is called checkpointing interval. When the system fails, the recovery process uses the log tape and the state saved at the most recent checkpoint to bring the system to the correct state that preceded the failure. This process is called error recovery and consists of loading the most recent state and then reprocessing all the activities, stored on the log tape, that took place since the most recent checkpoint and prior to failure.