云环境中软件性能异常诊断机制研究

发布时间：2018-06-16 11:31

本文选题：云计算 + 性能异常　；参考：《华中科技大学》2016年硕士论文

【摘要】：云服务软件需要对外提供不间断的在线服务,但是由于功能的复杂和代码规模的庞大使得软件中难以避免存在bug,如果这些bug引发了性能异常问题,开发者将很难对这些性能异常进行诊断。这些性能异常问题只有在特定的条件下才会出现,比如特殊的用户输入,系统配置等。在其他运行环境,将很难再次重现这些问题,已有的离线调试工具将无法使用。性能异常出现的时候往往不产生额外的错误信息,无法给开发者提供更多的帮助。因此,云环境中软件性能异常诊断机制研究具有重要意义。云环境中软件性能异常诊断系统旨在解决软件在线提供服务时出现的性能异常问题。首先,设计了系统调用划分机制,能够将软件运行时产生的系统调用划分为具有一定语义的执行单元;然后,设计了基于自组织映射模型自动建模方法,能够自动构建软件运行时的系统调用行为模型;最后,设计了基于距离的异常执行单元检测方法,以及基于多数票决的异常系统调用推断方法,能够准确捕捉系统调用行为的变化,通过比较不同执行单元之间的差异,检测出异常的执行单元,推断出性能异常的相关的系统调用,为开发者解决这个性能异常问题提供帮助。实验结果表明,云环境中软件性能异常诊断系统可以有效的诊断5个开源软件的实际性能bug,能够在平均7分钟之内推断出与性能异常最相关的系统调用。同时,它对测试的服务软件产生的运行时开销平均只有2.2%。
[Abstract]:Cloud service software needs to provide continuous online services, but because of the complexity of functions and the size of the code, it is difficult to avoid the existence of bugs in the software, if these bug cause abnormal performance problems, Developers will find it difficult to diagnose these performance exceptions. These performance anomalies occur only under specific conditions, such as special user input, system configuration, and so on. In other environments, it will be difficult to reproduce these problems again, and existing offline debugging tools will not be available. Performance exceptions often do not generate additional error messages and do not provide more help to developers. Therefore, it is of great significance to study the mechanism of software performance anomaly diagnosis in cloud environment. The purpose of software performance anomaly diagnosis system in cloud environment is to solve the problem of performance anomaly when software provides online service. Firstly, the partition mechanism of system call is designed, which can divide the system call generated by software runtime into execution units with certain semantics. Then, an automatic modeling method based on self-organization mapping model is designed. The system call behavior model of software runtime can be constructed automatically. Finally, the method of detecting exception execution unit based on distance and the method of extrapolation of exception system call based on majority vote are designed. Can accurately capture the system call behavior changes, by comparing the differences between different execution units, detect the exception of the execution unit, extrapolate the performance exception related system calls, Provides help for developers to solve this performance exception problem. The experimental results show that the software performance anomaly diagnosis system in the cloud environment can effectively diagnose the actual performance of five open source software bugs.It can infer the system calls most relevant to the abnormal performance in an average of 7 minutes. At the same time, it produces an average runtime overhead of only 2.2% for the service software being tested.
【学位授予单位】：华中科技大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP311.53

【参考文献】