随着云计算和大数据应用技术的发展,数据中心的数量和规模迅速发展,为了满足服务器之间大规模数据流动的需求,数据中心网络的通信能力面临巨大的挑战.传统数据中心中,网络的路由交换设备一般仅采用电域交换技术,电域交换技术虽然可以快速地、灵活地切换数据包的传输路径,但其本身存在通信带宽低、交换容量有限、高能耗等缺点.为了提高数据中心网络的性能、降低网络的能耗,最近的研究提出了若干基于慢速路径切换光器件的光电混合网络结构.它们通常只能将小部分数据量非常大的网络流放在高带宽的光网络上传输,其他的网络流仍然需要电域网络传输.随着快速可调波长激光器以及光波长路由器件的成熟,使光电混合网络结构灵活应对动态、多样的流量模式成为可能.该文基于快速可调波长激光器TWC(Tunable Wavelength Converters)和光波长路由器AWGR(Arrayed-Waveguide Grating Router),首次提出了一种OCS(Optical Circuit Switching)/EPS(Electrical Packet Switching)光电混合网络结构Ace-net.在文中详细描述了光电混合网络的结构以及带宽测量、仲裁控制、流量分配等机制,这些机制利用TWC器件快速波长变换的特性,能快速地应对网络流量的变化,使更多的网络流量在光域网络上传输;同时使用模拟器对此结构进行了评测,模拟结果表明此网络结构具有很好的网络性能.
With the increasing requirements of cloud services, the size and quantity of datacenters are growing rapidly. In order to meet the needs of large-scale data exchange among servers, datacenter network faces enormous challenges in the communication capabilities. In traditional data center, the routing and switching equipment typically uses electronic switching technology which is fast and flexible to switch packets between ports. But it inherently has low communication bandwidth and limited switching capacity. In order to improve the performance of datacenter network and reduce its energy consumption, many MEMS-based hybrid network architectures have been proposed recently. In these architectures, a fraction of flows which is very big is placed on the optical network while the rest flows are placed on the electronic network. With the maturity of Tunable Wavelength Converters (TWC) and Arrayed-Waveguide Gating Router (AWGR), the hybrid network has the opportunity to serve dynamic and various flows. In this paper, we propose Ace-net which is a hybrid OCS/EPS network architecture basing AWGR and TWC optical device. We first describe current research status of the hybrid network architecture in detail. Then we present the design and the key characteristics of Ace-net, which include traffic demand estimation, arbitration control and traffic distribution. Also, we evaluate this structure using our simulator and the simulation results show that the network structure has good performance.