NeTS: Medium: Collaborative Research: Flexible All-Wireless Inter-Rack Fabric for Datacenters Using Free-Space Optics

数据中心是支持我们社会的基础设施的重要组成部分。它们灵活地提供计算资源，使互联网经济成为可能。健壮的数据中心网络结构的设计具有挑战性，因为它必须满足几个目标，即高性能、低设备和管理成本、添加新服务器的增量可扩展性，以及其他实际问题，如布线复杂性、电源和冷却成本。该项目设想并提供了一种数据中心网络设计方法，与之前的架构完全不同：一种使用自由空间光学（FSO）通信链路的完全灵活的全无线结构，本质上是使用激光束通过空气无线传输数据。尽管室外FSO链路在恶劣天气条件下运行具有挑战性，但它们在温度控制数据中心环境中的使用，在信息处理能力方面提供了巨大的、尚未开发的潜力。所提出的网络设计为每个服务器机架配备了许多FSO设备，每个FSO设备都可以实时“操纵”，与另一个机架上的另一个FSO设备进行无线通信。转向能力使动态网络能够适应当前的网络流量。该项目的成功将通过提高IT服务在资本成本、运营费用和碳足迹方面的效率，在经济的几个关键部门产生明显的经济影响。该项目将把研究与教育结合起来，帮助学生成为数据中心网络行业的领域专家，并积极鼓励代表性不足的少数民族和妇女参与。该项目解决了在上述设想的背景下出现的一些科学挑战。(i)该项目将开发具有成本效益的FSO设备，这些设备具有较小的形状因素，可以在细粒度的时间尺度上进行控制。此外，该项目将开发可行的机制，以促进FSO链路的清晰视线，并解决实际操作挑战（例如灰尘、振动）。该项目将探讨设计灵活网络拓扑的算法基础；该项目将开发可扩展的网络管理解决方案，包括以最佳方式选择运行时拓扑和路由流量的算法；(v)在动态变化的拓扑结构中保证所需属性（例如，低拥塞和延迟）的机制；最后，该项目将建立一个概念验证系统原型，并在适当的平台上进行广泛评价。

Data centers are a critical piece of the infrastructure supporting our society. They flexibly supply the computing resources that enables the Internet-based economy. The design of a robust data center network fabric is challenging as it must satisfy several goals, viz., high performance, low equipment and management cost, incremental expandability to add new servers, and other practical concerns such as cabling complexity, and power and cooling costs. The project envisions and delivers on a datacenter network design approach that is radically different from prior architectures: a fully flexible, all-wireless fabric using Free-Space Optics (FSO) communication links, which essentially use laser beams to wirelessly transmit data through air. Although outdoor FSO links are challenging to operate under adverse weather conditions, their use in a temperature-controlled datacenter environments offers enormous and untapped potential in information-handling capacity. The proposed network design equips each rack of servers with a number of FSO devices, each of which can be "steered" in real-time to wirelessly communicate with another FSO device on a different rack. The steering ability enables a dynamic network that adapts to the prevailing network traffic. The project's success will have perceptible economic impact by making IT services more efficient - in terms of capital costs, operating expenses, and carbon footprint - across several critical sectors of the economy. The project will integrate the research with education to help students become domain experts in the datacenter networking industry, and actively encourage participation from under-represented minorities and women.The project addresses a number of scientific challenges that arise in the context of the above proposed vision. (i) The project will develop cost-effective FSO devices that have a small form-factor and can be steered at fine-grained timescales. (ii) In addition, the project will develop viable mechanisms to facilitate clear line-of-sight for the FSO links, and address practical operational challenges (e.g., dust, vibrations). (iii) The project will explore algorithmic foundations for the design of flexible network topologies; (iv) The project will develop scalable network management solutions including algorithms to optimally select a runtime topology and route traffic; (v) mechanisms to guarantee desired properties (e.g., low congestion and latency) in a dynamically changing topology; (vi) Finally, the project will build a proof-of-concept system prototype and conduct extensive evaluation on appropriate platforms.