Availability and performance evaluation of database systems under periodic checkpoints

Checkpointing roll back and recovery is a common technique to insure data integrity, to increase availability and to improve the performance of transaction oriented database systems. Parameters such as the the checkpointing frequency and the system load have an impact on the overall performance and it is important to develop accurate models of the system under study. We find expressions for the system availability and for the expected response time of the transactions from a model that, unlike previous analytical work, takes into account the dependency among the recovery times between two checkpoints. Furthermore, our model can incorporate details concerning the contention for the system resources.<>