Stochastic Performance Analysis of Phase Decomposition in Hyperledger Fabric

TL;DR

This paper develops a detailed measurement framework and stochastic performance model for Hyperledger Fabric, analyzing each phase's latency and scalability to guide deployment and optimize performance.

Contribution

It introduces a novel measurement framework for Hyperledger Fabric's individual phases and proposes a stochastic performance model validated on local and cloud clusters.

Findings

CPU cores linearly affect throughput

Raft ordering service scales well with nodes

Communication latency between client and service is significant

Abstract

Hyperledger Fabric is one of the most popular permissioned blockchain platforms. Although many existing works on the overall system performance of Hyperledger Fabric are available, a decomposition of each phase in Hyperledger Fabric remains to be explored. Admittedly, the overall system performance of Hyperledger Fabric might provide an end-user with satisfied performance information when invoking a transaction; however, it is far from informative when deploying a distributed system with specific performance goals, except for understanding each phase in Hyperledger Fabric. In this paper, we develop a measurement framework to characterize each phase's transaction and block data in Hyperledger Fabric based on the Fabric SDK Nodejs, where we thoroughly analyze and open source the implementation details of the measurement framework. We evaluate the performance of Hyperledger Fabric and have some interesting observations; 1. The number of CPU cores has a linear impact on the throughput of an endorsing peer. 2. The Raft-based ordering service shows good scalability with the number of ordering service nodes. 3. The communication latencies between the client and service in Hyperledger Fabric are significant. We then identify each phase's dominant latency in Hyperledger Fabric via primitive operation analysis and propose a stochastic computation model for performance analysis. We also use the alpha-beta communication model to analyze the corresponding communication latency. Finally, we validate the accuracy of the performance model on both local and cloud clusters. The experiment results and the performance model help guide the deployment of the Hyperledger Fabric service.