CAREER: Datacenter-Aware Local Storage Stacks

我们正在以前所未有的速度生成数据，其中大部分数据都存储在数据中心中。然而，支撑当前数据中心存储服务的存储堆栈基础设施是在前数据中心时代设计的，主要用于桌面和本地计算机。在数据中心系统中按原样使用时，这种存储堆栈会产生不必要的性能开销，资源利用率低下，并妨碍可持续性目标。因此，这将导致数百万依赖数据中心存储的用户的成本上升。该项目旨在设计和开发专门为数据中心量身定制的本地存储堆栈，以提高性能，优化资源利用，并为可持续发展目标做出积极贡献。关键的技术方法是将数据中心感知嵌入到本地存储堆栈中——特别是，数据中心系统被设计为具有固有冗余以容忍故障。该项目采用了广泛的视角，并确定了存储堆栈的四个基本支柱：性能、崩溃安全性、效率和数据完整性。在四个协同研究重点中，我们将通过利用内在冗余来重新设计和优化上述支柱背后的机制。这个项目将贡献新的技术、系统设计、协议和实际实现，以实现数据中心感知存储堆栈的愿景。该项目还非常重视教育，通过(i)设计存储/数据中心分布式系统的新课程，（ii）培训学生研究人员在存储硬件和软件上工作，以构建高效的数据中心存储，并最终将这些原则应用于其他未来领域，（iii）开发“口袋”集群教育工具包，为K-12学生提供真实的系统构建经验。这个项目将使所有的工件公开可用，包括出版物、数据集、代码和运行实验的说明。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。

We are generating data at unprecedented rates, and the bulk of this data is stored in datacenters today. However, the storage-stack infrastructure that underpins current datacenter storage services was designed in the pre-datacenter era, primarily for desktop and on-premise computers. When used as-is in datacenter systems, this storage stack incurs unnecessary performance overhead, poorly utilizes resources, and hampers sustainability objectives. Consequently, this results in elevated costs for millions of users relying on datacenter storage. This project aims to design and develop local storage stacks specifically tailored for datacenters to enhance performance, optimize resource utilization, and contribute positively to sustainability goals.The key technical approach is to ingrain datacenter-awareness into the local storage stack – in particular, the fact that datacenter systems are designed with inherent redundancy to tolerate failures. This project takes a broad view and identifies four fundamental pillars of storage stacks: performance, crash safety, efficiency, and data integrity. In four synergistic research thrusts, we will redesign and optimize the mechanisms underlying the above pillars by exploiting the inherent redundancy. This project will contribute novel techniques, system designs, protocols, and practical implementations to realize the vision of datacenter-aware storage stacks.This project also places significant emphasis on education by (i) designing new courses on storage/datacenter distributed systems, (ii) training student researchers to work on both storage hardware and software to build efficient datacenter storage and, eventually, apply the principles to other future domains, (iii) developing a "pocket" cluster educational kit to offer real systems building experience to K-12 students. This project will make all artifacts publicly available, including publications, data sets, code, and instructions to run experiments.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.