Collaborative Research: CNS Core: Medium: Reinvented Data Plane for Memory-Disaggregated Datacenters

云计算已经成为当今人类社会运作方式背后最关键的元素之一。云由大型数据中心提供支持，每个数据中心包含部署在全球各地的数千台机器。最近，资源分解被提出作为一种有效的手段来提高数据中心的可靠性和资源利用率，它将每种资源类型构建为独立的资源刀片，并将不同的资源刀片与网络结构连接起来。这个项目开发了一个新的内核/运行时数据平面，允许用户在内存分散的集群上高效地运行他们的应用程序。它建立在一个重要的洞察力之上：当前的内核/运行时模块之间没有任何协调，导致严重的性能问题。我们用四个主要的突破口来解决问题。前两个推力开发了支持语义的缓存、预取和交换模块，以便这些低级系统可以在考虑应用程序需要的情况下运行，从而提高效率。最后两个推力开发解聚感知的运行时和应用程序支持，使应用程序/运行时开发人员能够利用对象的位置信息来释放大量的优化机会。该项目构建了一个全新的内核/运行时数据平面，使未来资源分散的数据中心能够为用户应用程序提供卓越的效率。该项目的开发包括一名博士后学者以及几名研究生和本科生，其中包括来自少数群体的学生。它不仅提高了资源分解的技术水平，而且还培训了当前和未来的系统开发人员。该项目生成了一组构件，包括技术报告、发表的论文以及代码库。所有文物都可以在PIS的网站上在线获得。这一奖项反映了NSF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。

Cloud computing has become one of the most crucial elements behind how the human society operates today. The cloud is powered by large datacenters, each containing thousands of machines deployed across the globe. Recently, resource disaggregation has been proposed as an effective means to improve datacenter reliability and resource utilization by building each resource type as a standalone resource “blade” and connecting different resource blades with a network fabric. This project develops a novel kernel/runtime data plane that allows users to run their applications efficiently on memory-disaggregated clusters. It builds on an important insight: the current kernel/runtime modules do not have any coordination between themselves, leading to severe performance problems. We solve the problem with four major thrusts. The first two thrusts develop semantics-aware caching, prefetching, and swapping modules so that these low-level systems can operate with the consideration of what applications need, improving efficiency. The last two thrusts develop disaggregation-aware runtime and application support, making it possible for application/runtime developers to leverage objects' location information to unlock massive optimization opportunities. This project builds a brand-new kernel/runtime data plane, making it possible for future resource-disaggregated datacenters to deliver superior efficiency to user applications. The development of the project includes a postdoctoral scholar as well as several graduate and undergraduate students, including students from minority groups. It not only advances the state of the art of resource disaggregation, but also trains current and future system developers. The project produces a set of artifacts including technical reports, published papers, as well as code repositories. All artifacts are made available online at the PIs' websites.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.