Checkpointing and rollback-recovery algorithms in distributed systems

Abstract To keep it free of arbitrary failures, a distributed system may require taking checkpoints from time to time. In case of failures, the system will roll back to checkpoints where global consistency is preserved. Based on the concept of global consistency defined in this article, which eliminates both received-not-sent and sent-not-received types of inconsistencies, we developed a synchronous checkpointing algorithm C ∗ and an asynchronous roll back-recovering algorithm R ∗ with O ( VE ) message complexity, which has a relatively small coefficient, where V is the number of processors and E is the number of communication links in the system. Neither incoming message logging nor message-number calculation is needed. Our algorithms work well in complicated situations such as communication loops involving a minimal number of processors. Because of the feature of instant crash recovery and the overall low time, space, and message complexities, these algorithms are most feasible in real-time applications where rapid rollback recovery is crucial.