CSR: Small: Architectural and Operating System Support for Non-volatile Memory

自现代计算出现以来，用于构建计算机系统的技术已经稳定地划分为计算（操作数据），内存（在通电时保存数据）和存储（用于保留长期数据）。今天，一类新的技术正在出现，即非易失性存储器（NVM），它有望颠覆性地合并存储器和存储的概念。这项工作通过协同探索软件和硬件的变化，与一个由系统专家和几个重点架构师领导的团队，利用这种破坏的潜力。 这项工作开发了新的基准，并通过分析开发了新的设计，使未来的计算机系统能够以相同的成本完成更多的工作，或者由于更有效地访问NVM而以更低的成本完成相同的工作。此外，该项目还传播其NVM基准和软件工具，以进一步推动其他人的工作，并使NSF的资金更具影响力。更详细地说，工作开发和分析NVM基准与不同的根文件系统，持久性内存和数据存储。然后，它使用此分析来开发计算机NVM设计，以隐藏延迟，同时确保持久性，降低与多个内存控制器协调的成本，将通用图形处理单元（GPU）的想法转换为NVM使用，探索大型NVM的虚拟内存地址转换，并促进数据的快速原子发布。

Since the dawn of modern computing, the technologies used to build computer systems have stably partitioned into compute (manipulate data), memory (holds data while power is on), and storage (for retaining long-term data). Today a new class of technologies are emerging, namely non-volatile memory (NVM), that promises to disruptively merge the concepts of memory and storage. This work exploits the potential of this disruption by synergistically exploring software and hardware changes with a team led by a systems expert and architects with several foci. The work develops new benchmarks and- through analysis - new designs that enable future computer systems do more work at the same cost or the same work at less cost due to more effective access to NVM. Moreover, the project disseminates its NVM benchmarks and software tools to further the work of others and make NSF dollars more impactful. In more detail, the work develops and analyzes NVM benchmarks with dissimilar roots in file systems, persistent memory, and data stores. It then uses this analysis to develop computer NVM designs to hide latency while ensuring persistence, reduce the cost of coordinating with multiple memory controllers, transmutes ideas from general-purpose graphics processing units (GPUs) for NVM use, explores virtual memory address translation for vast NVMs, and facilitates fast atomic publication of data.