考虑失效恢复的组合服务性能建模与分析

Performance Modeling and Analysis for Composite Service Considering Failure Recovery

互联网环境下运行的组合服务易受到资源故障和组件失效影响而导致失效.已有的失效恢复措施在提高服务可用性的同时也会对服务的性能产生负面影响.为了对失效可恢复情况下的组合服务性能进行量化,通过综合组合服务失效类型和恢复策略,给出一种考虑失效恢复的组合服务性能分析模型.采用排队Petri网(queueing Petri net,简称QPN)描述组合服务的失效发生及其恢复处理过程,重点研究实施重试和替换策略的服务运行情况.详细描述了考虑失效恢复的服务节点和链路QPN模型的内部结构,在此基础上,通过服务交互机制构建组合服务分散执行的性能模型.最后,采用QPME工具仿真和比较不同失效发生率、失效类型分布和恢复策略下组合服务模型的性能表现.结果表明,该方法能够定量分析失效恢复对组合服务性能的影响,有助于指导不确定网络环境下的信息服务系统失效恢复策略实施方案的设计.

The Internet provides an open, dynamic, and uncertain environment. A composite service in this environment faces more failure threats caused by resource faults and component failures. The existing failure recovery measures can improve the availability of composite service and at the same time, bring the negative impact on service performance. In this paper, an analytical model considering failure recovery is proposed to quantitatively analyze the performance of composite service with the failure probability and recovery ability through synthesizes the fault taxonomy and recovery strategies of service failure. Queueing Petri nets （QPNs） are used to model the occurrence of failures and the corresponding recovery handling processes, which focuses on targeting the service executing process under retrying and replacement strategies. The inner structure of server node and link QPN models considering failure and recovery impact are expatiated and depicted. On this ground, a performance model of composite service with decentralized execution is given, and five interaction relations are taken into account. Finally, QPME tool is used to simulate and compare the performance measure results under different failure rates, failure styles proportions, and recovery strategies. The numerical results show that this method can quantitatively analyze the impact of failure recovery measure on service performance, and can also be helpful for providing rational solutions to implement failure recovery measure for information service system in uncertain network environments.