Yatch: Leaderless, Fault Tolerant Consensus Protocol

Abstract

Nowadays, with the advancement of computing power and faster internet, more and more applications are built where the machines are separated geographically apart, working together to give combined computing power faster than supercomputers and quick response time, better availability, and reliability. The machines have to coordinate to work together and provide coordination and agreement. Consensus protocols are used for coordination among geographically distant machines. The consensus protocols should be fast and simple. Protocols like Paxos, Raft, EPaxos, etc. which, solve the consensus problem in distributed systems. Generally, protocols are leader-based protocols that make them simpler, but leader machines can become the bottleneck in performance due to a single leader handling all communication. There are also leaderless protocols that solve the single leader problem but take more round trips. The number of roundtrips is an important criterion in distributed algorithms since it decides the speed and throughput of the algorithm. Distributed algorithms generally take more rounds in case of concurrent operations. This paper proposes a leaderless algorithm that takes two roundtrips in case of concurrent conflicting write operations. © 2022 IEEE.