Collaborative Research: CNS CORE: Medium: The Case for Blended Storage

近年来，数据存储技术的设计目标有两个：要么提高性能，要么降低每位成本。因此，在构建新的存储系统方面付出了巨大的努力，这些存储系统的组织方式要么非常快(但昂贵)，要么非常便宜(但速度慢)。该方案采用了一种不同的方法：它提出了混合存储系统，即结合多种不同存储技术的软件系统，以实现高性能和低比特成本。广泛使用的存储数据结构和算法是为单一存储介质设计的，当它们运行在具有不同特性的多个不同存储技术上时，会导致性能下降。因此，该提案重新考虑了支撑混合存储的现代存储系统的基本数据结构和算法，包括：索引、数据放置、日志记录、缓存、预取和近似成员结构。该方案将在一个专为混合存储设计的原型键值存储中设计和实现这些数据结构和算法。该研究将重新设计广泛使用的键值存储的数据结构和算法，以利用混合存储技术以低成本快速访问数据。由于Key-Value商店是许多云数据系统的基础，因此直接的好处将包括成本更低的高性能数据系统。这些反过来将转化为面向最终用户的经济高效的云服务。PI将在开源许可下发布软件构件，以支持其他研究人员，并将与硬件和云提供商合作，将研究转化为实践。该教育计划将把以存储为中心的系统设计提升为系统教育的一个组成部分。项目存储库将托管在https://github.com/princeton-sns的Github上，并无限期维护。该奖项反映了国家科学基金会的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

In recent years, data storage technology has been designed for one of two objectives: either improved performance or a lower cost per bit. Consequently, significant effort has gone into building new storage systems that are organized either to be very fast (but expensive) or very cheap (but slow). This proposal takes a different approach: it makes the case for blended storage systems, i.e., software systems that combine multiple different storage technologies, to achieve both high performance and low cost per bit.Widely-used storage data structures and algorithms are designed for a single storage medium in mind, which leads to poor performance when they are run on multiple different storage technologies with different characteristics. Therefore, the proposal rethinks the fundamental data structures and algorithms that underpin modern storage systems for blended storage, including: indexing, data placement, logging, caching, prefetching and approximate membership structures. The proposal will design and implement these data structures and algorithms in a prototype key-value store that is designed specifically for blended storage.The proposed research will redesign widely-used data structures and algorithms for key-value stores to take advantage of blended storage technologies for fast data access at a low cost. Since key-value stores underpin many cloud data systems, the immediate benefits will include high performing data systems at a reduced cost. These will in turn translate to cost-effective cloud services for end users. The PIs will release the software artifacts under open-source licenses to enable other researchers and will work with hardware and cloud providers to transition the research into practice. The education plan will elevate storage-centric systems design as an integral part of systems education.The project repository will be hosted on Github at https://github.com/princeton-sns and maintained indefinitely.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.