Dynamic Optimization for Trade-off in Hyperledger Fabric Towards Latency-Sensitive IoT Services

Hao Ding,Xuefeng Piao,Yuxi Hong,Yizhou Liu,Huihui Song,Meng Liu,Song Gao

DOI: https://doi.org/10.1109/icws62655.2024.00108

2024-01-01

Abstract:Hyperledger Fabric is currently the most widely used open-source enterprise blockchain platform in IoT services. However, Hyperledger Fabric still suffers from high latency in practical applications, hindering its further adoption in latency-sensitive IoT services. To address this issue, this paper selects Hyperledger Fabric v2.0 as the research target. The focus is on the latency of the ordering phase, which accounts for the highest proportion of latency in the EOV architecture. The ordering phase is primarily responsible for block packaging and generation, requiring careful parameter configuration to ensure the efficient operation of the blockchain system. We discuss the impact of BatchSize and BatchTimeout on block generation under two different consensus algorithms, Kafka and Raft. Additionally, based on the M/M/1 queue, we propose a latency analysis model for the ordering phase and analyze the causes of latency. Subsequently, we frame the optimization of blockchain experimentation as a Markov decision process, introducing a dynamic parameter configuration method grounded in deep reinforcement learning. This approach aims to strike a balance between BatchSize and BatchTimeout in response to network loads. The experimental results indicate that compared to existing methods, the proposed dynamic optimization strategy can reduce the average transaction latency of the blockchain system by 11.37%. Additionally, this strategy demonstrates good fault tolerance capability, with a reduction in the latency of node failures by approximately 6.35%. Therefore, the proposed dynamic optimization strategy meets the QoS requirements for IoT services.