中文摘要：

按需供给是PaaS平台面临的一个核心挑战。传统以局部优化为目标的反馈控制方法难以实现全局资源供给最优，为了更合理分配资源，利用性能模型预测Web系统的资源需求就变得至关重要。随着平台服务化的发展，其部署方式由封闭环境转为开放的服务形式，为模型的构造提出了新的挑战：由于应用和平台分属不同组织，大大增加了解系统全貌的难度，所以传统手工建模方法除了建模难度大之外，更难以在开放的环境下实施；在开放和动态的环境下，用户的使用难以预期，而用户行为的改变又会极大的影响系统行为，因而也有必要使模型适应用户行为的变化。针对上述问题，给出了一种动态性能建模工具的设计与实现。该工具在Web系统中插入必要的探针收集系统运行时状态，并输出为日志。通过周期性的对这些日志进行分析，从大量数据中提取出性能模型，使其与实际用户使用情况相符。文中以TPC-W基准测试为例，验证了该系统的有效性。

英文摘要：

On-demand provision is a core challenge for PaaS platform. Traditional feedback method solves the problem of local optimization and is hard to realize global optimization of resources provision. In order to allocate resources reasonably, performa- nce model plays an important role in predicting resources demand of Web applications. However, with the development of the service platform, the way of deployment is changed from closed environment to an opened one, such as PaaS. New challenges are broughr for performance modeling of such systems. First, it is difficult to build a performance model in advance, because applica- tions and platforms belong to different organizations. Second, in the opening and dynamic environment, the user behaviors is un- expected, which makes it is necessary to let the model adapt to the change of the mixed workload mode. To solve above prob- lems, a dynamic performance modeling tool is designed and implemented. The tool is used to collect the running condition of Web systems at first step by inserting necessary probes into those Web systems and writing to log files. And then through periodically analyzing these log files, the tool extracts a performance model from the mass data which is in coincidence with the use condition of Web users. At last, our work is evaluated with TPC-W bench mark, whose results can demonstrate the effectiveness of our approach.