An Adaptive Checkpointing Scheme for Distributed Databases with Mixed Types of Transactions

Recent study shows the possibility of a checkpointing mechanism that does not interfere with transaction processing, and yet achieves consistency of the checkpoints. The motivation for noninterfering checkpointing is to improve system availability. Noninterfering checkpointing mechanisms, however, may suffer from the fact that two separate database states need to be maintained by the system until all transactions, which are in progress when a checkpoint begins, come to completion. For distributed database systems in which data objects are partitioned across sites, checkpointing of this kind may not be practical if there exist many long-lived transactions. In this paper, we present a checkpointing algorithm that is noninterfering with transaction processing. It prevents the well-known “domino effect” and saves the intermediate results of a long-lived transaction in an adaptive manner, managing effectively both short and long-lived transactions in the system.