Using Common Knowledge to Improve Fixed-Dependency-After-Send

Abstract

Checkpoint patterns that enforce the rollbak-dependency trackability (RDT) property allow effcient solutions to the determination of consistent global checkpoints that include a given set of checkpoints. Fixed-Dependency-After-Send (FDAS) is a well-known RDT protocol that forces the dependency vector of a process to remain unchanged during a checkpoint interval after the first message-send event. In this paper, we explore processes' common knowledge about their behavior to derive a more effcient condition to induce checkpoints under FDAS. We consider that our approach can be used to improve other RDT checkpointing protocols.