CAREER: Automated, Portable, and Effective Application Guidance for Complex Memory Systems

具有多层不同性能和不同能力的异构存储器设备的计算机系统已经开始出现。传统的数据管理策略需要改变，以利用每一层中不同类型的内存，但这样做面临着许多挑战。该项目的主要目标是开发软件系统和工具，为新兴的存储系统提供更好的性能和效率，同时减少可能限制其采用的障碍。这项研究将深入了解如何收集和分析内存行为，典型的计算机应用程序如何使用内存资源，以及具有复杂内存硬件的系统应如何管理应用程序数据。这将使其他研究人员能够创建和探索更有效地管理内存的新技术，并有可能显着提高这些新兴系统的性能和能源效率。此外，这项工作有望在静态分析，程序分析和反馈导向优化领域产生新的知识，以预测和指导内存管理。这项研究的结果也将被整合到本科和研究生课程，以推进课程和学生的最新软件系统和内存技术的知识。推动这项研究的一个关键概念是，内存资源的分布和使用取决于垂直执行堆栈的不同层中发生的活动，包括应用程序，操作系统和硬件。通过加强这些活动之间的协调，这项研究将减少因使用有限信息进行数据管理决策而产生的效率低下，并实现细粒度，灵活和高效的数据管理。为实现这一目标，将建立一个协作数据管理框架，其中：1）应用运行时将使用自动化程序剖析和分析来根据数据对象的预期使用行为将数据对象划分到单独的堆区域中，并且将该信息传送到OS，和2）操作系统将合并并使用这些信息来指导物理内存分配和跨内存层层次结构的回收。与该项目相关的数据、代码和结果将通过公共git存储库提供。与相关资料库的链接将张贴在PI的网站上：http://web.eecs.utk.edu/~mrjantz/software.html。与项目相关的所有代码和数据将使用田纳西大学的中央文件系统进行存储和存档。该系统按照规定的时间表定期在现场备份，并在场外持续镜像。项目数据将在整个项目期间和项目结束日期后至少五年内保持和提供。该奖项反映了NSF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。

Computer systems with multiple tiers of heterogeneous memory devices with different performance and different capabilities have begun to emerge. Conventional data management strategies will need to be altered to take advantage of the different types of memory in each tier, but doing so faces a number of challenges. The primary goal of this project is to develop software systems and tools that enable better performance and efficiency for emerging memory systems while also reducing barriers that could limit their adoption. This research will develop insights into how to collect and analyze memory behavior, how typical computer applications use memory resources, and how systems with complex memory hardware should manage application data. This will allow other researchers to create and explore new techniques to manage memory more effectively, and has the potential to significantly improve both performance and energy efficiency in these emerging systems. Additionally, this work promises to generate new knowledge in the areas of static analysis, program profiling, and feedback-directed optimization for the purposes of predicting and guiding memory management. The results of this research will also be integrated into undergraduate and graduate coursework to advance curriculum and student knowledge of the latest software systems and memory technologies. A key concept driving this research is that the distribution and usage of memory resources depend upon activities that occur in different layers of the vertical execution stack, including the applications, operating system, and hardware. By increasing coordination among these activities, this research will reduce inefficiencies that arise from making data management decisions with limited information and enable fine-grained, flexible, and efficient data management. To achieve this objective, a collaborative data management framework will be created where: 1) the application runtime will use automated program profiling and analysis to partition data objects into separate heap regions according to their expected usage behavior and will convey this information to the OS, and 2) the OS will incorporate and use this information to direct physical memory allocation and recycling across the memory tier hierarchy.All of the generated data, code, and results related to this project will be made available through public git repositories. Links to the relevant repositories will be posted on the PI’s website: http://web.eecs.utk.edu/~mrjantz/software.html. All code and data related to the project will be stored and archived using the central file system at the University of Tennessee. This system is backed up regularly on-site following a defined schedule and continuously mirrored off-site. Project data will be maintained and made available throughout the duration of the project and for at least five years beyond the end date of the project.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.