基于动态插桩的程序分析工具的性能改进

摘要

Tools based on dynamic instrumentation are widely used in program analysis and software testing.However,these tools suffer from performance problems.The overhead of such tools consist of the overhead of dynamic instrumentation and the cost of user-defined analysis tools.In order to reduce the cost of user-defined analysis tools,this paper first analysed the compose of the overhead generated by these tools,then discussed some reason for these overhead and their influence on the tools with the help of experiment.According to the result of the analysis,it presented offline analysis to reduce the overhead,finally employed parallelized data collection to get the better performance.With the provided method,it reduces 5％～15 ％ CPU time occupied by the analysis tools.%基于动态插桩的工具被广泛应用于程序分析中,但该类工具都面临着严重的性能问题.这类工具的性能开销主要由两部分组成,即插桩引擎的开销和用户定义的分析程序的开销.为降低用户定义的分析工具的开销,首先分析了基于动态插桩的工具的性能开销的组成,并通过实验分析了造成性能开销的几点原因及其对工具的性能影响；根据分析结论提出了使用离线分析方式来优化工具性能,最后通过并行数据收集来进一步提升工具性能.使用该方法能减少分析程序5％～15％的CPU占用时间.