Systematic comparisons of RDT communication-induced checkpointing protocols

Rollback-dependency trackability (RDT) is a property stating that all rollback dependencies between local checkpoints are online trackable by using a transitive dependency vector. Since the RDT property was introduced, many communication-induced checkpointing protocols satisfying such a property have been proposed in the literature. Most protocols can be classified as three families, PCM family, EPSCM family and PMM family, according to their underlying RDT characterizations. Up to now, several theoretical analyses on comparing the performance of RDT protocols were addressed, but simulation studies on this topic are rare and not comprehensive. We present a simulation study for comparing RDT protocols systematically. The simulation is carried out in different computational environments. We will not only verify the results from existing theoretical analyses, but also explore the impact of optimizations on protocols. Our results can provide guidelines for understanding the efficiency of RDT protocols.