Improving ZooKeeper atomic broadcast performance by coin tossing

Abstract

ZooKeeper atomic broadcast (Zab) is at the core of ZooKeeper system, enforcing a total order on service requests that seek to modify the replicated service state. Since it is a leader based protocol, its performance degrades when the leader server is made to handle an increased message traffic. We address this concern by having the other, non-leader server replicas toss a coin and broadcast their acknowledgement of a leader’s proposal only if the toss results in an outcome of Head. We model the coin-tossing process and derive analytical expressions for estimating the coin’s probability of Head for a given arrival rate of service requests such that the dual objectives of performance gains and traffic reduction can be accomplished. Experiments compare the performance of our coin-tossing Zab version (ZabCT) with Zab performance and confirm that the dual objectives are demonstrably met under heavy workloads. Moreover, ZabCT meets all requirements essential for crash-tolerance provisions within Zab which can therefore be adopted in any ZabCT implementation.