基于流体随机Petri网的服务器集群再生模型  

Rejuvenation Model of Server Cluster with Fluid Stochastic Petri Net

作　　者：杜小智[1] 齐勇[1] 侯迪[1] 刘亮 

机构地区：[1]西安交通大学电子与信息工程学院,西安710049 [2]IBM中国研究院,北京100094

出　　处：《计算机科学》2009年第7期131-134,共4页Computer Science

基　　金：国家863项目(2006AA01Z101);IBM合作项目资助

摘　　要：服务器集群是提高系统QoS和可用性的一种方法,但由于需要长期连续运行提供服务,集群系统仍然会存在软件老化现象,从而造成系统的处理能力随时间的增加而降低、失效率随时间的增加而增加。现同时考虑软件老化对系统处理能力与失效率的影响,采用流体随机Petri网对集群系统的工作模式进行建模,给出系统各个状态概率的数值分析方法,选择系统的可用性和吞吐量作为评价集群系统性能的指标,并进行仿真试验。结果表明该方法能很好地描述系统且具有可扩展性,分析并求得系统的最佳再生周期。Server clustering is an usual method to improve QoS and availability of system. But due to the need for longterm continuous operation to provide services, there exists software aging phenomenon in cluster system, which makes the system service rate is reduced and the failure rate is increased with the time increases. In this paper, the effect of software aging and the effect of failure were all considered, and fluid stochastic Petri net was adopted to model the working pattern of cluster system. Then the analysis methods of probability of each system state were given, and the formulas of system availability and throughout were presented. The availability and throughout were taken as the performance metrics of the cluster system. And some numeric experiments were done. The results show that the method introduced in this paper has the ability to describe system and is extendable. By using this method, the optimal rejuvenation period is gotten.

关 键 词：软件再生 软件老化 流体随机PETRI网 集群 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]