LowPaxos: State Machine Replication for Low Resource Settings

Abstract

State Machine Replication (SMR) is a popular framework for building highly available fault-tolerant systems, and widely uses distributed consensus as an implementation approach. Consensus protocols usually require a leader replica to coordinate the actions of other members in ensuring that the sequential command log to retrieve and update the state of the system is consistent despite failures. These protocols often assume homogeneity of the deployment environment and resource capabilities of all replicas each with an equal chance of leadership. However, challenged environments are mostly heterogeneous and the choice of a good leader can offer better performance. This paper introduces LowPaxos as a distributed consensus protocol for challenged environments. LowPaxos uses the computing and network capabilities of the replicas, and operational environment to designate the best leader, and adapt accordingly to changes characteristic of these settings. LowPaxos is evaluated against both leader-based (MultiPaxos) and leaderless (EPaxos) protocols and demonstrates performance gains of upto 5X and 2X respectively in a heterogeneous challenged setting.

Keywords

• State
• replication
• resource-constrained
• challenged
• low resource
• consensus