A Roll-Forward Recovery Scheme for Solving the Problem of Coasting Forward for Distributed Systems

Abstract

In this paper, a new index-based hybrid checkpointing scheme has been proposed to tackle the problem arising due to "coasting forward". This scheme uses both incremental and communication-induced checkpoints. Failures have been classified as hard and soft in order to take the advantage (i.e. possible reduction of rollback) which the incremental checkpoints offer, while designing the recovery approaches for these two types of failures. The presented theoretical results show that for soft failures, the recovery approach offers maximum roll-forward.