Efficient Diskless Checkpointing and Log Based Recovery Schemes

Checkpointing and message logging are the popular and general-purpose tools for providing fault tolerance in distributed systems. Diskless checkpointing schemes enable frequent checkpointing without a performance penalty. The present work extends James S Plankʹs Diskless checkpointing scheme (N+1 Parity) by introducing ʹTimeoutʹ mechanism to checkpoint programs with high locality of reference. This mechanism enables applications with high locality of reference to take checkpoints periodically. The limitation of N+1 Parity scheme is that all the processes freeze their respective computation, while taking synchronous checkpoints. The proposed scheme solves this problem by introducing a new message logging technique namely partial message logging which allows asynchronous checkpointing at both sender and receiver. Correctness of the scheme is established through a set of proofs. This paper includes the performance evaluation of proposed scheme by making use of distributed simulator test-bed. The results indicate that proposed scheme outperforms N+1 Parity Scheme.