一种基于ε-greedy的领导选举方法  

A Leader Election Method Based on ε -greedy Strategy

作　　者：许津铭 蔡亮[1] 孙路 尹可挺[1] XU Jinming;CAI Liang;SUN Lu;YIN Keting(College of Computer Science and Technology,Zhejiang University,Hangzhou 310013,China)

机构地区：[1]浙江大学计算机科学与技术学院,浙江杭州310013

出　　处：《无线电工程》2024年第4期826-834,共9页Radio Engineering

摘　　要：随着区块链的广泛部署,无人协同等延迟敏感型的应用对区块链系统的低时延需求日益提高。在协同场景下,区块链节点通常跨地域部署,节点异构性较强。在基于领导节点的拜占庭容错(Byzantine Foult Tolerant, BFT)共识协议中,不稳定的或能力较差的领导节点将导致不必要的高延迟,并降低区块链的可用性,特别是在资源有限的移动或传感器网络下。针对上述问题,提出了ε-LE,一种带有网络感知的领导选举方法,基于节点到领导节点的通信延迟测量结果,采用ε-greedy策略对领导节点进行选择,使得当前性能较优或网络中关键位置的节点具有更高概率成为领导节点,从而优化共识延迟。相较于AWARE等方法,ε-LE实现O(N)的通信复杂度,更加适用于具备线性通信复杂度的共识协议。实验结果表明,ε-LE能够选择可优化集群共识延迟的节点作为领导节点,在线性拓扑网络中实现了约21%的吞吐量提升。With the widespread deployment of blockchain,latency-sensitive applications such as unmanned collaboration have increasing demands for low latency of blockchain.In collaborative scenarios,blockchain nodes are usually highly heterogeneous and deployed across regions.In leader-based Byzantine Fault Tolerant(BFT)consensus protocols,a weak and unstable leader will lead to unnecessarily high latency and reduce the availability of blockchain,especially under resource-limited mobile or sensor networks.To solve the problem,ε-LE,a leader election method with latency awareness,is proposed.Based on the measured communication delay from the node to the leader,this method uses anε-greedy strategy to select the leader.As a result,there will be a higher probability for stable nodes in critical positions to become the leader,thus optimizing the consensus latency.Compared with methods such as AWARE,ε-LE achieves O(N)communication complexity and is more suitable for consensus protocols with linear communication complexity.The experiment shows that the critical replica that may optimize consensus latency will be selected byε-LE as the leader,and an approximately 21%throughput improvement in a linear topology network is achieved.

关 键 词：领导选举 共识 拜占庭容错 延迟感知 

分 类 号：TP319[自动化与计算机技术—计算机软件与理论]