Traffic Control for RDMA-Enabled Data Center Networks: A Survey
Traffic Control for RDMA-Enabled Data Center Networks: A Survey
复制标题
DOI：
10.1109/jsyst.2019.2936519
复制
发表时间：
2020-03
影响因子：
4.4
通讯作者：
Zehua Guo;Sen Liu;Zhi-Li Zhang
中科院分区：
文献类型：
--
作者：
Zehua Guo;Sen Liu;Zhi-Li Zhang
数据中心是云计算的基础架构，已在世界范围内广泛部署，以适应不断增长的云计算需求。数据中心网络（DCN）连接数据中心中的数十或数十万个服务器，并使用流量控制方案来启用服务器之间的数据传输。云中的各种新应用都提出了有关DCN的流量控制的新要求，例如低延迟和高吞吐量。 DCN中的现有交通控制计划遭受了复杂的内核处理，无法满足要求。远程直接内存访问（RDMA）绕过内核处理以使快速内存在网络上移动，被认为是有前途的解决方案。在本文中，我们介绍了有关传统RDMA，传统DCN和启用RDMA的DCN的交通管制方案的调查，并解释了它们的局限性。我们还将现有方案与拥塞控制，性能和组件区分开来。为了鼓励未来的研究，我们指出了这项研究的一些潜在研究方向。
Data centers, the infrastructure of cloud computing, have been widely deployed around the world to accommodate the increasing cloud computing demands. A data center network (DCN) connects tens or hundreds of thousands of servers in the data center and uses a traffic control scheme to enable the data transmission among servers. Various new applications in cloud present new requirements on traffic control of DCNs, such as low latency and high throughput. Existing traffic control schemes in DCNs suffer from the complicated kernel processing and cannot satisfy the requirements. Remote direct memory access (RDMA), which bypasses the kernel processing to enable fast memory moving across a network, is recognized as a promising solution. In this article, we present a survey of traffic control schemes for traditional RDMA, traditional DCNs, and RDMA-enabled DCNs and explain their limitations. We also differentiate the existing schemes from congestion control, performance, and components. In order to encourage future research, we point out some potential research directions of this research.