Collaborative Research: SHF: Medium: Reinventing Fuzz Testing for Data and Compute Intensive Systems

新兴的数据密集型和计算密集型软件应用程序的重要性继续以越来越快的速度增长。云计算框架通过提供现成的资源，使开发和运行大数据应用程序变得更加容易。计算机体系结构的最新趋势包括异构和专门化以提高性能，例如基于fpga的硬件加速器。虽然模糊测试已经成为传统软件应用程序中检测正确性和性能缺陷的一种有效技术，但由于其较长的延迟，它不容易适用于数据密集型和计算密集型应用程序。现在，这样的数据和计算密集型应用程序正被集成到关键任务系统中，它们的健壮性和正确性是最重要的。该研究为大数据应用和异构应用领域带来了自动化模糊测试的成功。这项研究产生了一套开源测试工具，以提高整个应用程序的质量，转化为资源、时间和成本节约。它有三个创新的组成部分：(1)针对数据密集型应用的新的输入突变技术和测试延迟减少方法，(2)针对异构计算应用的有效指导度量和反馈监控方法，以及(3)新的性能感知模糊策略，以诱导数据倾斜、计算倾斜和内存倾斜。本研究旨在使在数据密集型和计算密集型领域编写软件的软件工程师、数据科学家和定量分析人员受益。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

The importance of emerging data-intensive and compute-intensive software applications continues to grow at an increasing rate. Cloud-computing frameworks make it easier to develop and run big data applications by providing readily available resources. Recent trends in computer architectures incorporate heterogeneity and specialization to improve performance, such as hardware accelerators built on FPGAs. While fuzz testing has emerged as an effective technique for detecting correctness and performance defects in traditional software applications, it is not easily applicable to data-intensive and compute-intensive applications due to their long latency. Now that such data- and compute-intensive applications are being integrated into mission-critical systems, their robustness and correctness are of the highest priority.This research brings the success of automated fuzz testing to the domain of big data applications and heterogeneous applications. This research is producing a suite of open-source testing tools to improve overall application quality, translating into resource, time, and cost savings. It has three innovative components: (1) new input-mutation techniques and testing latency reduction methods for data-intensive applications, (2) effective guidance metrics and feedback-monitoring methods for heterogeneous computing applications, and (3) new performance-aware fuzzing strategies to induce data skews, compute skews, and memory skews. This research aims to benefit software engineers, data scientists, and quantitative analysts who write software in data-intensive and compute-intensive domains.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.