Performance Improvement by Using Pipelined Execution on Hyperledger Fabric

Abstract

The rapid growth of proofs of concept blockchain applications leads to increasing interest in understanding and improving blockchain performance at scale. However, the lower performance of blockchain restricts its application in some fields. Our work is focused on evaluating and improving the performance of Hyperledger Fabric, which is the most popular blockchain platform for enterprises. In previous works, the major bottleneck incurred in the validation & commit (V&C) module was studied, and many performance issues arising with it were alleviated to some context. The throughput is still only 900 transactions/second in our experiment. In this paper, a comprehensive latency evaluation for the V&C module was first performed. Then, according to the analysis of the evaluation results, a pipelined execution technology was proposed to process multiple blocks in parallel. Additionally, some pipeline acceleration schemes were also proposed to further improve the performance. Our experiments indicated performance improvements of 4.38× for LevelDB and at least 2× for CouchDB. Notably, our optimizations are transparent to blockchain applications and are suitable for integrating into a future version of Fabric.