基于分布式环境一种动态依赖关系发现的方法

本文提出了一种为分布式应用动态构造依赖性模型的方法。这个方法通过对系统进行主动干扰来获得建模的先验知识,然后基于贝叶斯网络构造方法,对分布式应用的组件间关系建立依赖性模型。和传统的被动建模技术不同的是,这种主动方法不需要事先对系统细节充分了解,它通过在运行环境中部署探针,捕捉和测量与部署的主动干扰相关的系统反馈,通过机器学习的方法识别分布式应用中构件间的动态调用的依赖关系,为分布式应用建立动态运行过程中的依赖性模型。动态建立的依赖性模型可用于分布式应用的运行时管理,用于分布式应用执行过程中的故障定位和恢复,对于分布式应用自主计算环境的实现,提供一种实用的方法。

An active approach for dependency discovery in a distributed environment

This paper presents a method for dynamically building dependency models in distributed environments. The method relies on active perturbation to the system to obtain prior knowledge for building the model, and then constructs the dependency model based on the methods of building Bayesian networks. Unlike more traditional passive techniques, our active approach requires little initial knowledge of the implementation details of the system. It actively perturbs system components while deploying the probes to monitor the system' s response, and then employs the method of machine- learning to identify the dynamic dependencies between system components and construct a dynamic dependency model for the distributed application. Our approach to dynamically construct depen- dency model, is useful for the management of distributed applications on running time. And it can be used for problem determination and recovery during the execute phase of distributed applications. It provides a practical approach for the realization of automatic computing in distributed environments.