A Data Flow Framework with High Throughput and Low Latency for Permissioned Blockchains

Zhenxing Hu,Shengjie Guan,Wenbo Xu,Zhen Xiao,Jie Shi,Pengze Li,Qiuyu Ding,Hui Ding,Chao Zeng

DOI: https://doi.org/10.1109/ICDCS57875.2023.00088

2023-01-01

Abstract:In permissioned blockchains, the bandwidth of consensus nodes is mainly consumed by transaction ordering and block distribution; hence, the allocation of consensus nodes' bandwidth makes a significant difference to the system throughput. Previous research focuses on the consensus layer and attempts to optimize consensus protocols to improve throughput, which, however, neglects the impact of data distribution on the throughput and transfers performance bottlenecks to the network layer. In fact, the overall throughput of permissioned blockchains is co-determined by data production in the consensus layer and data distribution in the network layer. This paper proposes a novel data flow framework composed of Predis and Multi-Zone. The former is a data production strategy for permissioned blockchains that employ leader-based BFT protocols and the latter, its corresponding network topology. Predis enables each consensus node to contribute its idle bandwidth for block content pre-distribution so that a much higher volume of transactions can be confirmed in one consensus round, significantly increasing consensus efficiency. Multi-Zone is a network topology to distribute blocks. It can regulate the bandwidth consumption of consensus nodes at a certain value during data distribution and effectively reduce block propagation latency. To test our framework, we implement Predis based on Hotstuff and PBFT, respectively, and experiments show that Predis significantly improves their throughput by 300% to 800%. Multi-Zone is implemented on BFT-SMaRt and compared with random and star network topologies, and it is shown that Multi-Zone holds excellent scalability and the capability of reducing block propagation latency by at least 50%.