一种分布式并行服务器节点故障检测算法  

A Node Fault Detection Algorithm in Distributed Parallel Server

作　　者：左朝树[1] 刘心松[2] 邱元杰[2] 陈小辉[2] 李可[2] 

机构地区：[1]中国电子科技集团公司第30研究所,成都610041 [2]电子科技大学计算机科学与工程学院,成都610054

出　　处：《电子科技大学学报》2007年第1期119-121,125,共4页Journal of University of Electronic Science and Technology of China

摘　　要：故障检测技术是实现分布式并行服务器容错的基础。为了尽可能减小故障检测算法对系统通信开销的增加,寄生式自适应故障检测算法被提了出来。该算法依靠系统内部固有的信息交换进行故障检测,而通过自回归AR模型预测消息的传输时间和处理时间,并以此自动调整故障检测的阈值,达到自适应系统运行状况的目的,最后通过伪代码的形式描述了该算法的实现。该算法已被应用于分布式并行数据库系统DPSQL,较好地实现了节点故障检测。Fault detection technology is the base of fault-tolerance in distributed parallel server. To reduce the communication cost that fault detection algorithm brings to the greatest extent, Autoecious Adaptive Fault Detection （A2FD） algorithm is proposed. Depending on the exchange of inherent information processed by system, the algorithm carries out fault detection. To achieve the goal that it adapts itself to system status, the algorithm adjusts the key value of fault detection according to transmission and transaction time of the message predicted by Auto Regression （AR） model. Finally, the realization of the algorithm is described with the pseudocode. The algorithm has been applied to distributed parallel database system--DPSQL, which has realized node fault diction well.

关 键 词：分布式并行服务器 故障检测 自适应 寄生 

分 类 号：TP302.8[自动化与计算机技术—计算机系统结构]