NeTS: Small: Merging Traffic for Energy Conservation in Enterprise & Datacenter Networks

如今，计算和通信设备所消耗的能源已成为开展业务的一大成本。事实上，随着我们转向更高速的网络，并在世界各地增加互联网的部署，这一成本预计将大幅增长。部署的基础设施的规模和能源成本的主要贡献者包括企业网络和数据中心网络。企业网络部署在工作场所，并被证明利用率严重不足，峰值利用率通常只有容量的一小部分。同样，数据中心网络在相当长的时间内也被证明负载非常轻。在这两种情况下，与实际传输的数据相比，网络基础设施消耗的能量非常高。这样做的原因有两个。首先，网络交换机设计为即使在没有流量的情况下也始终以峰值速率运行；其次，交换机端口将一对一映射到终端系统，即使这些系统以可用容量的一小部分来加载端口。这项研究项目的结果是对网络交换机进行了根本性的重新设计，既保留了用户体验，又实现了显著的节能。这项工作的关键思想是开发一种名为合并网络的硬件设备，它位于交换机和相连的系统之间。进入交换机的流量被合并在一起，并馈送到数量较少的端口，从而允许将端口密度较大的交换机替换为较小的交换机，或者关闭交换机的大部分电源，如线路卡和内存条。虽然这个想法的好处是显而易见的，但实现它的挑战很多。在硬件层面，挑战在于如何以最低的能源成本和数据丢失来合并流量。本项目采用的方法是使用带有数字控制的模拟组件来构建合并网络，以管理第二层协议问题。现代交换简化了一套预期交换机端口到终端系统的1-1映射的第2层协议。通过使用合并网络打破这种映射，实际上会导致不正确的协议行为。通过在每台交换机中利用端口虚拟化来解决此问题。因此，第2层协议不受影响，同时网络的能源使用量大幅降低。该项目使用原型、实验和模拟来全面描述企业和数据中心网络中的这种新的交换机架构。基于Click模块化路由器的实验系统将被插入到大学网络中，以测量实际应用中的能量节约以及了解这种重新设计对整体网络行为的副作用。在数据中心网络和大学级企业网络上收集的流量统计数据将被用于模拟器中，以在更大范围内分析协议行为。这个项目具有变革性，因为它重新设计了互联网的基本构件，从而实现了惊人的节能。这项工作的更广泛影响范围从从根本上重新设计互联网的关键组件(交换机)到直接影响通信基础设施中的能源使用。这对行业的影响是巨大的，因为这里的工作将影响未来高速交换机的设计，在这种情况下，合并网络将被整合到架构中，而不是作为外部附加组件。在该项目期间收集的交通统计数据将提供给研究界，从而提供急需的企业级追踪来源。研究界还将受益于在Click软件基础上实施IEEE 802.1协议。该项目将通过开发一个新的班级、让学生参与测量和开发实验平台来加强对未来劳动力的培训。

Energy consumed by computing and communication devices has become a significant cost ofconducting business today. Indeed, as we move to higher speed networks and increase thedeployment of the Internet to all corners of the world, this cost is expected to grow dramatically.Major contributors to the size and energy cost of the deployed infrastructure include enterprisenetworks and data center networks. Enterprise networks are deployed in the workplace and havebeen shown to be heavily under-utilized with peak utilizations typically a fraction of capacity.Similarly, data center networks have also been shown to be very lightly loaded for significantlengths of time. In both these cases, the energy consumed by the networking infrastructure is veryhigh, in relation to the actual data transmitted. The reasons for this are twofold. First, networkswitches are designed to always run at peak rates even in the absence of traffic and second, switchports map one-to-one to end systems even though these systems load the ports at a tiny fraction ofavailable capacity. The outcome of this research project is a fundamental re-design of the networkswitch in a way that preserves user experience but results in dramatic energy savings.The key idea in this work is the development of a hardware device called a merge network, whichsits between the switch and connected systems. Traffic coming into the switch is merged togetherand fed to a smaller number of ports allowing either the replacement of large port-densityswitches with smaller switches or the powering off of large portions of the switch such as linecards and memory banks. While the benefit of this idea is evident, the challenges in realizing itare many. At the hardware level, the challenge is in merging traffic with minimal energy cost anddata loss. The approach adopted in this project is to build the merge network using analogcomponents with a digital control for managing Layer 2 protocol issues. Modern switchesimplement a suite of Layer 2 protocols that expect a 1-1 mapping of switch ports to end-systems.By breaking this mapping using a merge network, one effectively causes incorrect protocolbehavior. This issue is addressed by utilizing port virtualization within each switch. Thus, theLayer 2 protocols remain unaffected while the network sees tremendous reduction in energyusage. The project uses prototyping, experimentation and simulation to fully characterize this newswitch architecture in enterprise and data center networks. Experimental systems based on ClickModular Router will be inserted into the College network to measure the energy savings in realapplication as well as to understand side effects of such a redesign on overall network behavior.Traffic statistics collected at a data center network and the University-level enterprise networkwill be utilized in a simulator to analyze protocol behavior at a larger scale. This project istransformative in that it redesigns the basic building block of the Internet resulting in dramaticenergy savings.The broader impact of this work ranges from fundamentally redesigning the key component ofthe Internet (the switch) to directly influencing the energy usage in the communicationsinfrastructure. The impact on industry is significant in that the work here will influence the designof future high-speed switches where the merge network is incorporated into the architecturerather than as an external add on. Traffic statistics collected during this project will be madeavailable to the research community and will thus provide a much-needed source of enterprise-leveltraces. The research community will also benefit by the implementation of IEEE 802.1protocols into the Click software base. The project will enhance the training of the futureworkforce via the development of a new class and including students in hands on measurementand development of the experimental platform.