ITR/SY: Blurring the Line between OSes and Storage Devices

长期以来，磁盘系统一直是性能和管理问题的根源，而且随着时间的推移，它们的重要性也在增长。 特别值得关注的是缓慢改善机械定位延迟。 为了解决这些问题，存储设备和主机系统有着丰富的资源和知识. 设备固件和操作系统（OS）软件的工程师各自独立地积极利用他们的本地知识和资源来缓解磁盘性能问题。本研究的总体目标是增加这两组工程师之间的合作，这将增加端到-本研究将深入了解可用的设备和工作负载信息，以及如何在真实的系统中利用这种理解将需要更有表现力的接口和新的合作算法。 需要探讨的两个具体示例是：（1）设备端专业化，例如自由块调度，可以访问系统提供的有关优先级和即将到来的需求的信息，以及（2）操作系统端专业化，例如轨道对齐的扩展，可以访问设备提供的有关物理数据布局和固件缓存算法的详细信息。 此外，这项研究产生的见解将有助于存储系统中的端到端教育和思考。

Disk systems have long been a source of performance and managementproblems, and they grow in importance as the years pass. Of particularconcern are the slow-to-improve mechanical positioning delays. Toaddress these problems, storage devices and host systems have abundantresources and knowledge. Independently, engineers of both devicefirmware and operating system (OS) software aggressively utilize theirlocal knowledge and resources to mitigate disk performance problems.The overall goal of this research is to increase the cooperationbetween these two sets of engineers, which will increase theend-to-end performance and robustness of computer systems.This research will develop a deep understanding of available deviceand workload information and how it can be collectively exploited.Exploiting this understanding in real systems will require moreexpressive interfaces and new cooperative algorithms. Two concreteexamples to be explored are: (1) device-side specializations, suchas freeblock scheduling, that have access to system-providedinformation about priority levels and forthcoming demands, and(2) OS-side specializations, such as track-aligned extents, thathave access to device-provided details about physical data layoutand firmware cache algorithms. In addition, the insights generatedby this research will assist end-to-end education and thinking instorage systems.