CRII: CSR: System Techniques to Exploit the Byte-Accessibility of Solid-State Drives

基于闪存的固态硬盘（SSD）已经成为现代存储系统的支柱，以支持越来越多的数据密集型应用程序，如图形分析和数据库。几十年来，系统软件和应用程序使用通用块接口访问ssd中的数据。今天，通过内存映射的输入/输出接口以字节和块粒度访问ssd是可行的。然而，目前还不清楚字节可寻址的ssd将如何集成到今天的内存存储层次结构中，它们将如何影响系统软件（如文件系统）的设计和实现，以及应用程序将如何从ssd的字节可访问属性中受益。本计划拟研究在现代记忆体储存层级中，支援及利用ssd的位元组存取能力的系统技术。首先，该项目建议重新考虑统一的内存-存储层次结构，并有效地使用字节寻址SSD作为主存储器的一部分，以简化双字节和块可访问接口的管理和可编程性。其次，该项目提出了自适应页面迁移机制，使应用程序能够同时透明地从字节可寻址的大规模SSD和快速动态随机存取存储器（DRAM）中获益。第三，该项目提出了一种新的细粒度数据持久化抽象，它将重新思考数据持久化机制的设计和实现，并进一步减少相关的性能开销和对存储设备的冗余写操作。随着ssd在各种计算平台上的广泛应用，所提出的研究活动将有利于许多具有高性能持久存储需求的研究领域，包括科学计算、大数据分析和金融服务。该项目还将通过产业合作为技术转化为实践开辟一条道路，并促进本科生和代表性不足的学生的研究活动。这个项目将把研究进展和结果整合到一个以内存和存储技术为中心的新课程中。由此产生的软件、数据集、出版物和课程材料将通过GitHub: https://github.com/sprlab/uniflash上的开源存储库发布，并在本项目完成后至少维护五年。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。

Flash-based solid-state drive (SSD) has been the backbone of modern storage systems to support a growing number of data-intensive applications such as graph analytics and databases. For decades, systems software and applications used a generic block interface to access data in SSDs. Today it is feasible to access SSDs in both byte and block granularity through memory mapped input/output interfaces. However, it is still unclear how byte-addressable SSDs will be integrated in today's memory-storage hierarchy, how will they affect the design and implementation of systems software such as file systems, and how applications will benefit from the byte-accessible property of SSDs.This project proposes to investigate the systems techniques to support and exploit the byte-accessibility of SSDs in modern memory-storage hierarchy. First, the project proposes to rethink the unified memory-storage hierarchy and efficiently use byte-addressable SSD as part of the main memory to ease the management and programmability of the dual byte and block-accessible interfaces. Second, the project proposes adaptive page migration mechanisms to enable applications to gain benefits from both byte-addressable, large-scale SSD and fast dynamic random-access memory (DRAM) concurrently and transparently. Third, the project proposes a new abstraction of fine-grained data persistence which will rethink the design and implementation of data persistency mechanisms and further reduce the associated performance overhead and redundant writes to storage devices. As SSDs have been widely deployed in various computing platforms, the proposed research activities will benefit many fields of study with high-performance persistent storage requirements including scientific computing, big data analytics, and financial services. The project will also carve a path for technology transition to practice through industrial collaborations, as well as facilitate research activities for undergraduate and underrepresented students. This project will integrate the research developments and results into a new course that centers around memory and storage technologies. The resulting software, dataset, publications, and course materials will be released through an open-source repository on GitHub: https://github.com/sprlab/uniflash and maintained for at least five years beyond the completion of this 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.