RDT-Partner: An Ef£cient Checkpointing Protocol that Enforces Rollback-Dependency Trackability

Checkpoint patterns that enforce rollback-dependency trackability (RDT) have only on-line trackable checkpoint dependencies and allow ef£cient solutions to the determination of consistent global checkpoints. The design of RDT checkpointing protocols that are ef£cient both in terms of the number of forced checkpoints and in terms of the data structures propagated by the processes is a very interesting research topic. Fixed-DependencyAfter-Send (FDAS) is an RDT protocol based only on vector clocks, but that takes a high number of forced checkpoints. The protocol proposed by Baldoni, Helary, Mostefaoui and Raynal (BHMR) takes less forced checkpoints than FDAS, but requires the propagation of an matrix of booleans. In this paper, we introduce a new RDT protocol, called RDT-Partner, in which a process can save forced checkpoints in comparison to FDAS during checkpoint intervals in which the communication is bound to a pair of processes; a very interesting optimization in the context of client-server applications. Although the data structures required by the proposed protocol maintain the complexity of FDAS, theoretical and simulation studies show that it takes virtually the same number of forced checkpoints than BHMR.