CAREER: Combating Performance Bugs in Software Systems

在主流软件中，性能缺陷会导致意外的性能下降和能源浪费。随着多核时代对单线程性能的提高越来越少，以及对能效的日益重视，这些bug很容易使生产软件的速度降低数倍，并对软件质量构成越来越大的威胁。以往针对性能bug的工作，包括测试和分析，大多将软件视为黑盒，并且通常对性能bug的损害预防得太晚或无效。这个NSF CAREER研究项目旨在开发系统和工具，这些系统和工具采用白盒方法来解决性能错误的公开问题。首先，它进行了实证研究，以丰富对现实世界中的性能错误的理解。其次，它开发了可以在bug出现之前、期间和之后自动检测性能bug的工具。第三，设计了一个测试框架，在软件发布前暴露性能缺陷。最后，本研究亦探讨多执行绪程式中的效能错误问题，本研究的教育重点是扩阔软件课程，让学生对软件，特别是多执行绪软件的效能和正确性问题有更多的认识和练习，本研究将增进对软件效能浪费问题的了解，提供实质的工具支援，以协助降低软件开发和效能测试的成本，并通过更快的软件改善软件用户的日常体验。它还将有助于提高生产运行软件的能源效率并保护现场环境。

Performance bugs cause unintended performance degradation and energy waste in mainstream software. These bugs can easily slow down production software for several times and are posing increasing threat to software quality with the meager increases in single threaded performance in multi-core era and an increasing emphasis on energy efficiency.Previous works for performance bugs, including testing and profiling, mostly treat software as a black-box, and are usually too late or ineffective in preventing the damage of performance bugs. This NSF CAREER research project seeks to develop systems and tools that take a white-box approach to addressing the open problem of performance bugs. First, it conducts an empirical study to enrich the understanding of performance bugs in real-world. Second, it develops tools that can automatically detect performance bugs before, during, and after bug manifestation. Third, it designs a testing framework to expose performance bugs before software release. Finally, it also looks at the performance bug issue in multi-threaded programs.The education focus of this project is to broaden the software curriculum to bring students more awareness and exercise of performance and correctness issues in software, especially multi-threaded software.This research will improve the understanding of performance waste problem in software, provide substantial tool support to help lower software development and performance testing costs, and improve software users' everyday experience through faster software. It will also help improve the energy efficiency of production-run software and protect environment in the field.