Tuning Paxos for High-Throughput with Batching and Pipelining

Abstract

Paxos is probably the most popular state machine replication protocol. Two optimizations that can greatly improve its performance are batching and pipelining. Nevertheless, tuning these two optimizations to achieve high-throughput can be challenging, as their effectiveness depends on many parameters like the network latency and bandwidth, the speed of the nodes, and the properties of the application. We address this question, by first presenting an analytical model of the performance of Paxos that can be used to obtain values for tuning batching and pipelining. We then present results of experiments validating the model and investigating how these two optimizations interact in a WAN. Results for LAN are also mentioned. The results show that although batching by itself is usually sufficient to maximize the throughput in a LAN environment, in a WAN it must be complemented with pipelining.