CSR: Small: Exploiting the Potential for Fine-Grained Memory Management via a Virtual Memory Framework Enhanced by Page Overlays

CSR：小：通过页面覆盖增强的虚拟内存框架挖掘细粒度内存管理的潜力

• 批准号: 1618595
• 负责人: Todd Mowry
• 金额: $ 49.97万
• 依托单位: Carnegie-Mellon University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2016
• 资助国家: 美国
• 起止时间: 2016-09-01 至 2020-08-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1618595&HistoricalAwards=false
• 关键词: CSR; Small; Exploiting; Potential; Fine

Many areas of science and technology have experienced transformational changes in recent years as a result of computing on extremely large data sets. To help enable future breakthroughs in science and technology, we also need breakthroughs in the ability of computers to efficiently perform data-intensive computations. Perhaps one of the most important inventions in the history of operating systems and computer hardware was the creation of virtual memory support. Virtual memory support not only fulfilled its original motivation of giving the operating system the flexibility to manage the hardware memory resources in a way that is convenient for programmers, it also provided unexpected benefits, such as enabling the 'copy-on-write' mechanism that dramatically reduces the effective cost of duplicating large amounts of memory. While virtual memory is a powerful tool, it has one major shortcoming: it provides flexibility at only a relatively coarse granularity (e.g., 8KB "pages" of contiguous memory). This research is exploring how a new technique for eliminating this shortcoming can enable a number of new fundamental mechanisms for performing data-intensive computations much more efficiently.The key insight in this project is that through a modest extension of the hardware that supports virtual memory (called 'page overlays'), the benefits of virtual memory can be provided at a much finer granularity: e.g., 32B rather than 8KB. This orders-of-magnitude improvement in precision does not simply make existing virtual-memory-enabled optimizations more efficient: it enables fundamental new optimizations that had previously been impractical. For example, preliminary work has demonstrated that this new page overlays mechanism can dramatically improve the performance of the 'copy-on-write' optimization by enabling a new (much more efficient) 'overlay-on-write' mechanism. This project is exploring how page overlays can enable six additional fundamental new mechanisms for making data-intensive computations faster and more energy-efficient. With these six new mechanisms, future computer systems should be better suited to the increasingly data-intensive workloads of the future.

近年来，由于对超大数据集的计算，许多科学技术领域都经历了变革。 为了帮助实现未来科学技术的突破，我们还需要突破计算机有效执行数据密集型计算的能力。 也许操作系统和计算机硬件历史上最重要的发明之一就是虚拟内存支持的创建。 虚拟内存支持不仅实现了其最初的动机，即让操作系统能够以方便程序员的方式灵活地管理硬件内存资源，而且还提供了意想不到的好处，例如启用“写时复制”机制极大地降低了复制大量内存的有效成本。虽然虚拟内存是一种强大的工具，但它有一个主要缺点：它仅在相对粗粒度（例如，连续内存的 8KB“页面”）上提供灵活性。 这项研究正在探索一种消除这一缺点的新技术如何能够启用许多新的基本机制来更有效地执行数据密集型计算。该项目的关键见解是通过对支持虚拟内存的硬件进行适度扩展（称为“页面覆盖”），虚拟内存的好处可以以更细的粒度提供：例如，32B 而不是 8KB。 这种数量级的精度改进不仅仅使现有的虚拟内存优化更加高效：它还实现了以前不切实际的基本新优化。 例如，初步工作表明，这种新的页面覆盖机制可以通过启用新的（效率更高）“写入时覆盖”机制来显着提高“写入时复制”优化的性能。 该项目正在探索页面覆盖如何启用六种额外的基本新机制，从而使数据密集型计算更快、更节能。 有了这六种新机制，未来的计算机系统应该能够更好地适应未来日益增长的数据密集型工作负载。