一种基于临界区的分布式决策算法  

A Distributed Decision-making Algorithm Based on Critical Sections

作　　者：邢隽炜 石连星 白聃 XING Junwei;SHI Lianxing;BAI Ran(Children’s Hospital of Fudan University,Shanghai 201102,China;Shanghai Xiaoyun Info Tech Co.,Ltd.,Shanghai 201109,China;Shanghai Fengxian District Central Hospital,Shanghai 201406,China)

机构地区：[1]复旦大学附属儿科医院,上海201102 [2]上海霄云信息科技有限公司,上海201109 [3]上海市奉贤区中心医院,上海201406

出　　处：《微型电脑应用》2025年第2期191-193,202,共4页Microcomputer Applications

摘　　要：分布式系统中各服务节点集群状态的一致性是保证系统行为正确性的基础。目前大多数分布式存储系统采用中心化的管理服务来维护集群状态,为各节点提供集群状态查询服务,以确保各节点获取的集群信息的一致性,避免多节点执行分布式事务时可能出现的并发冲突问题。然而,中心化方法适用于实时性要求较低的集群状态维护,在实时性要求较高的场景中会降低分布式服务的并发执行效率。因此,提出了基于临界区的分布式决策算法。在临界区之内,采用中心化决策和分布式执行的方式完成任务;在临界区之外,各节点独立决策并执行任务。所提算法结合分布式决策和集中化管理的优势,极大限度地保证了分布式服务执行的并发效率和正确性。实验结果表明,与完全中心化决策的方式相比,所提算法可以将分布式并发执行效率提升30~50个百分点。In a distributed system,the consistency consensus among service nodes regarding the cluster state is the foundation for ensuring the correctness of the system behavior.Currently,most distributed storage systems use centralized management services to maintain the cluster state,provide state query services to ensure consistency in the information obtained by each node.This helps to avoid concurrent conflicts that may occur when multiple nodes execute distributed transactions.However,centralized methods are suitable for maintaining cluster state with low real-time requirements.In scenarios with high real-time requirements,the methods reduce the efficiency of concurrent execution of distributed services.Therefore,this paper proposes a distributed decision-making algorithm based on critical sections.In the critical section,tasks are completed using centralized decision-making and distributed execution.Outside the critical section,each node independently decides and executes tasks.The proposed algorithm combines the advantages of distributed decision-making and centralized management,maximizes the concurrent execution efficiency and correctness of distributed services.Experimental results show that compared to fully centralized decision-making methods,the proposed algorithm can improve distributed concurrent execution efficiency by 30~50 percentage points.

关 键 词：分布式系统 中心化集群管理 分布式共识 分布式决策 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]