This paper proposes a reusable software dynamic analysis methodology for componentized Internet applications, which is composed of two important parts: execution model construction and runtime behavior analysis. Firstly, the related concepts of behavior observation are presented. Secondly, according to the behavior observations extracted from running applications, the multi benchmarking behavior model and runtime component interaction model are constructed. Then, by means of the multi benchmarking behavior model, a software behavior analysis method is proposed to detect functional and performance anomaly, and to locate the aging or failure components. Finally, using the component interaction model, a failure propagation algorithm is proposed to localize the failure. The proposed models and methods have been applied to JPetStore which is a typical J2EE application. The experimental results show that it is effective, and a good solution for locating failure component and localizing failure in software rejuvenation decision and recovery phases.