Collaborative Research: CNS Core: Medium: Movement of Computation and Data in Splitkernel-disaggregated, Data-intensive Systems

资源分解有望改变我们设计和运营云基础设施的方式。但是，尽管它对数据中心的运营和成本效益产生了积极的影响，但分解带来了开销，因为每次访问远程内存、存储或加速器时都要引入PCIe和网络往返——对于数据密集型应用程序来说，这种开销尤其明显。本提案将回答的核心研究问题是，分解的出现对数据密集型系统设计者几十年来所做的架构决策意味着什么，同样，需要哪些操作系统级原语来使这些系统适应分解的设置。为此，本工作提出了一种用于数据密集型系统的分解操作系统和中间件的协同设计，以实现高性能和易用性。像本工作所针对的数据密集型系统这样的数据密集型系统为当今世界的大部分领域提供了动力，为从商业信息到互联网搜索和医疗保健的所有内容提供了分析。拟议的工作将帮助数据中心运营商将其基础设施转换为支持资源分解，同时最大限度地减少用通用框架编写的应用程序的性能开销。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。

Resource disaggregation promises to change the way in which we design and operate cloud infrastructure. But despite its positive impact on the operation and cost-efficiency of data centers, disaggregation comes with overheads that arise from introducing a PCIe and network round trip for each access to remote memory, storage, or accelerators -- overheads that are particularly pronounced for data-intensive applications. The core research question that this proposal will answer is what the advent of disaggregation implies for the decades of architectural decisions made by the designers of data-intensive systems, and likewise, what OS-level primitives are required to adapt these systems to a disaggregated setting. To that end, this work proposes a co-design of disaggregated operating systems and middleware for data-intensive systems that achieves both high performance and ease of use, Data-intensive systems like the ones targeted by this work power much of today’s world, providing analytics for everything from business informatics to Internet search and healthcare. The proposed work will assist data center operators in transitioning their infrastructure to support resource disaggregation while minimizing performance overheads to applications written in common frameworks.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.