End-to-end congestion control mechanisms in 5G networks: from device to data center

数据中心通常建立在以太网络上，以受益于低成本的以太网设备。然而，以太网在数据包丢失和延迟方面存在严重的流量工程能力不足，而在密集数据中心整合服务器需要仔细的资源管理。另一方面，云计算与移动自组织网络在5G中的结合，将通过支持大量具有突发数据的不同带宽需求的应用，如实时搜索和数据分析，显然将构建新的主流网络。此外，5G网络预计将提供成倍增长的容量、更低的延迟、无处不在的连接以及更高的可靠性和可用性。数据中心流量的不稳定特性使5G需求具有相当大的挑战性，需要设计创新的拥塞控制机制。因此，最终目标是为5G网络提出从数据中心终端主机到终端用户的端到端拥塞控制解决方案。UQAM-NSERC-Prompt魁北克合作项目的目标是通过在对商用交换机硬件进行最小修改的情况下提出创新的拥塞控制方案，提高数据中心主机之间的网络带宽共享。因此，我们建议研究适合5G网络范式的不同拥塞控制方法。我们提出了涉及不同数据中心组件(主机、交换机)的新的以太网交换矩阵拥塞控制方案的设计和研究。我们还提出了传输层增强，以构建5G网络中全面的端到端拥塞控制解决方案。此外，为了评估我们提出的拥塞控制方案，我们使用了严格的性能评估方法。事实上，我们将构建一个网络规模的数据中心网络模拟器，用于第一阶段的性能评估。之后，将开发一个基于Linux的网络仿真器来执行更真实的测试和实验。此外，我们提出的涉及交换机的修改将在现场可编程门阵列环境中实现和测试。

Data centers are typically built on Ethernet networks to benefit from low cost of Ethernet devices. However, Ethernet suffers from crucial traffic engineering capability shortage in terms of data packet loss and latency while consolidating servers in dense data centers needs careful resource management. On the other hand, the combination of cloud computing with mobile ad-hoc networks in 5G will clearly build new prevailing networks by supporting a large number of different bandwidth-hungry applications with bursting data, such as real-time search and data analysis. Furthermore, 5G networks are expected to provide exponentially more capacity, lower latency, ubiquitous connectivity, as well as increased reliability and availability. The erratic property of data center traffic makes 5G requirements quite challenging and necessitates the design of innovative congestion control mechanisms. Hence, the ultimate goal is to propose an end-to-end congestion control solution for 5G networks, from data center end-host to end-users. The objective of this UQAM-NSERC-Prompt Québec collaborative project between is to improve network bandwidth sharing between hosts in data centers, by proposing innovative congestion control schemes with minimal modifications of commodity switches hardware. Thus, we are proposing to investigate different congestion control approaches seemly to the 5G networks paradigm. We propose the design and the study of new Ethernet Fabric congestion control schemes involving different components of data centers (Hosts, switches). We are also proposing transport layer enhancement to build a comprehensive end-to-end congestion control in 5G networks solution. Furthermore, to assess our proposed congestion control schemes, we are using rigorous performance evaluation methods. Indeed, we will build a web-scale data center network simulator to be used at the first phase of performance evaluation. Afterward, a Linux-based network emulator will be developed to perform more realistic tests and experimentations. Also, our proposed modification involving the switch will be implemented and tested in a FPGA environment.