The Research On Methods Of Performance Analysis And Diagnosis For Server Software

With the rapid development of Internet, Web-based applications have been impregnated into our work and life. Many traditional information systems have been transplanted into the Internet, and new applications modes, such as E-business, are emerging. Enterprises and persons are in favor of Web-based applications because they are extensive, interactive, fast and easy-to-use. Performance is an important factor of a successful Web-based application. Reducing the defects and improving the quality of Web server software have become important issues of software test.At present, the tools and methods in performance testing have become mature and are used widely in Web sever software. However, after a performance issue is discovered, there is no good tool or method to diagnose it. In fact, many achievements have been acquired in performance measurement, analysis, and diagnose of Web-based applications. Some simple performance issues such as insufficient memory can be found by these methods. However, they cannot be used for joint diagnosis of multi-performance issues. In this paper, we investigated the methods for analyzing and diagnosing the performance of Web server software. By mining the Web log data, we can acquire performance measurement indices, and then define the performance issue mode, which is the basis of performance analysis. The methods proposed in this paper can not only find performance issues, but also accurately locate them. These methods improve the degree of automation in performance analysis and diagnosis. The research includes the following three aspects:(1)Building a netty indice for performance measurement.Acquiring and analyzing Web log data are very important in analyzing the performance of software on Web server. However, current tools are confined to processing Web log data and counting the Web browsing and access modes, which cannot effectively support the performance analysis and diagnosis. Based on the environment and features of Web server software, we researched the methods to acquire and analyze multilevel and mass log data in terms of data mining, such as Web server log, operating system log and performance counter log. We also designed the performance index systems aimed at performance diagnosis. These methods are universal and can effectively support the subsequent performance analysis and diagnosis.(2)Proposing the innovative concept of performance defect mode.We also analyzed some common performance defects associated with Web-based applications, summarized their features, and proposed the concept of performance defect mode. After analyzing the influencing factors and evaluation standards, we built the representing model of the performance defect mode and defined the set of performance index and constraints between them. To a certain degree, the concept of performance defect mode assures the accuracy and improves the efficiency of performance analysis and diagnosis.(3)Building a diagnosis model for performance defect.Based on the features of every performance defect mode and their computational methods, we analyzed the interdependency between performance indices, built a diagnosis model for the performance defect, and designed the algorithms for analyzing and diagnosing performance defects. The efficiency and efficacy was tested by a set of comprehensive experiments. This model will improve the performance and stability of Web-based applications.The research of performance issues of server software would provide scientific methods, which could be used to solve performace testing of server software. This research result would play an important role to improve the degree of informatization in E-business, E-government and financial service.