无应答严次序发布传播复制策略算法及Java实现  

Algorithms and Java Implementation of NASO Publish-Propagate Replication Scheme

作　　者：刘晖[1] 彭勤科[2] 沈钧毅[1] 燕彩蓉[1] 

机构地区：[1]西安交通大学软件研究所,陕西西安710049 [2]西安交通大学系统研究所,陕西西安710049

出　　处：《小型微型计算机系统》2004年第7期1307-1312,共6页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目 (60 175 0 15 )资助

摘　　要：介绍无应答严次序发布传播复制策略的定义、准则和算法 ,并给出 Java实现中数据表及核心对象的定义、线程与 Connection间的封锁关系速查表 .该策略在集群服务器节点中容错 4地复制各种文件、操作和数据更新 .其中 ,双钥锁允许加锁者或下一个申请者释放互斥锁 ,次序管理遵循四个准则避免乱序、意图歪曲、因果牵制和反向因果牵制四个数据不一致问题 ,全局 /局部This paper presents the definitions, rules and algorithms of No Acknowledgement Strict Order Publish Propagate Replication (NASOPPR) scheme, which is a relaxed, update everywhere replication scheme. Some Java implementation experiences, including the definitions of tables, descriptions of key objects and a quick lookup table for the synchronization relationships between threads and database connections, are also provided. NASOPPR scheme is a fault tolerance scheme used to replicate all kinds of files, operations and data updates to the given replicas composing a cluster. It is an integration of OOP, IP Multicast, twin keys locking and sequence management tech. Twin keys Locking allows either the lock holder or the next applicant to unlock an exclusive lock. Sequence management abides four rules to avoid four inconsistency problems, i.e., divergence, intention violation, causality diversion and inverse causality diversion. Global/Local/Status Order Numbers that act as timestamp also can be used to detect and recover the data loss or data error of IP Multicast and such a process could be deferred until update is to occur.

关 键 词：集群 数据复制 互斥锁 串行化 数据一致性 

分 类 号：TP311.133[自动化与计算机技术—计算机软件与理论]