应用粒度及性能分析.docx

摘要摘要 摘要 摘要 随着Java EE应用系统规模的扩大，如何对系统资源消耗进行定量测量和定 性分析，以便充分合理利用计算资源，是业界和学术界共同关注的重要课题。 论文研究Java EE应用系统在不同粒度下的资源消耗模式。 论文介绍了常用Java EE框架及其运行流程，提出了按照粒度级别划分应用 系统的概念，将Java EE应用分为服务器、逻辑单元及组件三个粒度级别，并设 计和实现了相应的性能监测方法。 论文以学位与研究生教育发展评估平台为研究对象，在服务器粒度级别用 JMeter监测不同并发度下系统的响应时间，用DTrace监测不同请求的整体资源 消耗。在逻辑单元粒度级别，通过hotspot探针识别不同逻辑单元的包前缀的方 法标记当前逻辑单并与资源消耗数据相关联，得到不同逻辑单元的资源消耗数 据。在组件粒度级别，通过筛选调用堆栈的方法获得关键步骤的函数类名及方 法名，利用函数切换点输出的时间戳结合HTML5技术绘制各组件调用的时序 图。进而将资源消耗数据与时间相关联，得到了各组件资源消耗的时空分布。 论文提出的粒度划分概念、对资源消耗进行性能监测的方法及得到的结果对类 似系统的实现和部署具有重要的指导作用。 关键词：Java EE应用粒度性能研究DTrace JMeter AbstractWith Abstract With the expansion of Java EE application system，both industry and academia have shown great concern with how to quantitatively measure and qualitative analysis system I'eSOUl'Ce consumption in order to optimize the usage of computing resources． This paper studied the resource consumption pattern of Java EE application system in different granularity． Commonly used Java EE frameworks and tlleir workflow are introduced in this paper．The paper proposed the concept of granularity based partition of application system and divide Java EE application into three granularity level Server Level， Logic Part Level and Component Level．Meanwhile，performance monitoring methods are designed and implemented at different level． The Academic Degree＆Graduate Education Evaluating Platform is set勰the object of the study．JMeter is used to monitor response time in different concurrency． DTraee is used to monitor the OVer all resource consumption of different requests in Server Level．On Logic Part Level，different prefix of logic part packages are detected and marked by hotspot probe．Resource consumption data in different logic parts are gained by making conjunction of logic part marker and resource consumption data．On Component Level，filtered invocation stack is used to capture class name and method name of key functions．Combining the function switching timestamp and HTML5 drawing technique， an invocation sequence diagram i