NSF-BSF: SHF: CCF: Small: Collaborative Research: Hardware/Software Design of Durable Data Structures and Algorithms for Non-Volatile Main Memory

如果内存在断电（即崩溃）时不丢失其内容，则称为非易失性内存。用于非易失性存储器的硬件正在快速发展，预计不久将在许多体系结构中取代易失性DRAM作为主存储器。该项目需要开发并发数据结构和软件，这些数据结构和软件可以利用新兴形式的非易失性存储器，提供简单、更快、更节能的崩溃恢复。这项研究有可能通过增强现代社会日益依赖的系统的鲁棒性和容错性来造福社会。这项研究的结果也可能显著提高能源效率，因为这样的系统将需要更小的缓存和更少的耗能的二级存储设备。新的课程开发活动也计划作为这个项目的一部分。本项目研究如何使常见的数据结构（如队列、跳过列表、散列映射等）具有持久性。考虑了不同级别的持久性，特别强调高性能和易于编程。该项目还评估了几种内存架构，包括传统的多层内存和新兴的近数据处理架构。另一项补充工作是研究刷新缓存以及在易失性和非易失性内存区域之间移动数据所需的硬件原语。在这些努力的基础上，该项目将产生一系列性能和节能的持久并发数据结构，这些数据结构打包为软件库，允许程序员轻松地进行调整。评估将在综合基准测试以及并发应用程序（如并发键值存储、偷取工作调度器和并发内存管理）上进行。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。

Memory is said to be non-volatile if it does not lose its contents when it loses power (i.e., crashes). Hardware for non-volatile memory is advancing fast, and soon it is expected to displace volatile DRAM for main memory in many architectures. This project entails developing concurrent data structures and software that can exploit emerging forms of non-volatile memory that provide simple, faster, and more energy-efficient crash recovery. This research has the potential to benefit society by enhancing the robustness and fault-tolerance of systems on which modern society increasingly depends. The results of this research may also significantly improve energy-efficiency, since such systems would require smaller caches and fewer accesses to energy-hungry secondary storage devices. New curriculum development activities are also planned as part of this project. This project investigates how to make common data structures, such as queues, skip-lists, hash maps, and others, durable. Different levels of durability are considered, with special emphasis on high performance and ease of programming. The project also evaluates several memory architectures, including both conventional multilevel memory and emerging near-data processing architectures. A complementary effort investigates hardware primitives needed to flush caches, and to move data between volatile and non-volatile memory regions. Building on these efforts, the project will produce a collection of performance and energy-efficient durable concurrent data structures, packaged as software libraries that allow for easy adaptation by programmers. Evaluation will be done on synthetic benchmarks as well as concurrent applications such as concurrent key-value stores, work-stealing schedulers, and concurrent memory management.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.