Efficient incremental checkpoint algorithm for primary-backup replication

Abstract

Replication protocols are widely used for enabling fault tolerance and reliability features in distributed systems aiming fast recovery and seamless transition. In this study, we propose an efficient incremental checkpoint algorithm for primary-backup replication protocols to increase the system throughput. We developed an in-memory key-value store configured by the primary-backup replication protocol and set it up on the geographically distributed nodes of the PlanetLab overlay network. We performed measurements for metrics of interest on both the client and the primary replica side. Our findings show that the proposed incremental checkpoint algorithm not only assures 2-3 times lower average blocking times but also guarantees a near-steady minimum average blocking time.