Pipelining and Overlapping: Techniques to Improve Both Throughput and Latency in BFT Consensus Blockchain

Abstract

For decentralized characteristics of blockchain platforms, Byzantine Fault Tolerance (BFT) consensus protocols have been widely used to reach an agreement among nodes in the presence of malicious node attacks. However, due to their communication complexity, BFT consensus protocols are considered one major cause of the limited scalability of a blockchain platform. In this study, we propose two performance improvement techniques — Pipelining and Overlapping — to simplify the BFT consensus protocol without sacrificing decentralization. These techniques are designed to concurrently optimize throughput and minimize latency, irrespective of an increase or broad dispersion in consensus node deployment. We demonstrate the efficacy of these techniques through their application to AuditChain, a private blockchain leveraging a PBFT-like consensus mechanism with linear communication complexity. In a WAN environment utilizing AWS EC2 with 4 consensus nodes deployed in one region (ap-northeast-2), the system achieves about 1997 tx/s with a latency of 0.1 seconds. By further enhancing the distribution characteristics of our experimental environment, even with 32 consensus nodes distributed across two regions (ap-northeast-2, us-east-1), the system still achieves about 1995 tx/s with a latency of 0.3 seconds. These findings suggest that the proposed techniques are capable of significantly improving the performance metrics of BFT-based blockchain systems.

Keywords

• Blockchain
• Byzantine fault tolerance
• consensus protocol
• performance