NeTS: Small: Collaborative Research: Ultrascale WDM-based Datacenter Networks: Architecture Design and Control Algorithms

庞大的数据中心日益成为互联网的核心。该项目将为数据中心开发一个基于光纤的网络架构，以及相关的网络控制算法，该架构高度可配置、可扩展，可满足未来数据中心的要求。提出的体系结构是分层的，在分层结构的最低层使用低成本的无源光纤，在较高层使用电子和可配置光交换的组合，从而在光学系统和设备的成本和能力之间取得平衡。此外，无源光纤的使用显著降低了能源消耗，这已成为大型数据中心的重要考虑因素。我们的研究计划包括体系结构开发和算法设计，包括：1.开发一种基于波分复用(WDM)的低成本、可扩展、可配置的DCN体系结构，以满足未来数据中心的通信需求。开发动态重构算法，根据流量状况的变化改变网络拓扑结构，从而降低网络的通信负荷。调查无源光纤的使用以降低能耗，并在能耗方面比较不同的架构备选方案。建议的架构有可能显著改变DCN的设计方式，显著提高其可扩展性、健壮性和能源效率。此外，这项调查将包括：(1)通过期刊和会议出版物传播结果，以促进该领域的发展；(2)将通过这项工作开发的新理论和算法纳入高级研究生班；(3)为本科生开发实验项目；(4)向行业转移技术(包括可能的行业合作伙伴早期采用)和(5)对研究生和本科生的培训。

Huge data centers increasingly form the heart of the Internet. This project will develop an optical-fiber-based network architecture for data centers, together with associated network control algorithms, that is highly configurable, scalable, and meets the requirements of future data centers. The proposed architecture is hierarchical and uses low-cost passive optics at the lowest levels of the hierarchy, and a combination of electronic and configurable optical switching at higher layers, thus striking a balance between cost and capability of optical systems and devices. Moreover, the use of passive optics significantly reduces energy consumption, which has become an important consideration in large data centers. Our research plan includes both architecture development and algorithm design, including:1. Develop a Wave-Division-Multiplexing (WDM)-based DCN architecture that is low-cost, scalable, and configurable, for meeting the communication requirements of future data centers.2. Develop dynamic reconfiguration algorithms for changing the network topology in response to changes in traffic conditions, in order to reduce the communication load on the network.3. Investigate the use of passive optics in order to reduce energy consumption, and compare different architectural alternatives in terms of energy consumption.The proposed architecture has the potential to dramatically change the way that DCNs are designed, significantly improving their scalability, robustness and energy efficiency. In addition, this investigation will include: (1) dissemination of the results through journal and conference publications to advance the field; (2) incorporation of new theory and algorithms developed through this work into advanced graduate classes; (3) development of experimental projects for undergraduate students; (4) technology transfer to industry (including potential early adoption by an industry partner) and (5) training of graduate and undergraduate students.