Research On Network Traffic Scheduling Of Photoelectric Hybrid Data Center Based On SDN

In the era of cloud computing,large number of new increasing cloud applications has posed huge challenges to traditional data center networks.As the demand of computing capacity increases for cloud services,the network size continues to expand,exposing the disadvantages of high energy consumption,small capacity,and low biesction bandwidth in traditional data centers.In order to overcome these shortcomings of traditional data center networks,researchers introduced optoelectronic hybrid interconnects and all-optical network interconnects into data center networks.Some of these networks use static configuration methods,which are simple to configure but cannot dynamically establish and teardown optical paths to adapt to traffic variations,leading to network throughput and resource utilization.Others use one-level optical switching networks like OXC,which directly establish optical paths between N input and output ports,but are limited by the number of ports of the optical switching module,making it difficult to meet demands of the dynamic optical path establishment between thousands of top-of-rack switches in data centers,and scalability is not good enough.Therefore,a new data center architecture and traffic routing mechanism need to be designed to fulfill the traffic scheduling and routing requirements of data centers,which aims to improve network throughput and scalability performance.In this thesis,a data center network with electronic packet switching and optical switch network based on Clos-network-based optical switching network is proposed,and a hierarchical traffic scheduling strategy in this architecture is designed.The architecture also uses software-defined network technology,which collects traffic information by Open Flow switch,transmits it to the controller of the software-defined network,and then the controller finds the traffic rout.The proposed architecture leverages both electronic switching network and optical switching network to implement traffic scheduling.In the optical switching network,a three-level Clos cross structure is used,in which multiple small-scale optical switching modules are cascaded to realize cross-connection among a large number of ports in the data center.Direct optical paths are established at the source and destination ports of large flows to transmit large flows directly,while the electric switching network is used for the transmission of small flows,which reduces network congestion in the data center and improves network throughput.The thesis analyzes the cost and energy consumption of the proposed architecture.The results show that with the expansion of the data center network,the proposed network architecture can effectively reduce the cost and energy consumption of the data center network,comparing with the traditional tree network,Fat-tree,C-Through and Helios.Ryu and Mininet are used to simulate the traffic scheduling of a data center network,in which the throughput and delay of the proposed network architecture and other architectures are compared.The simulation results show that the throughput of the proposed architecture is increased by 17.51% and the delay is decreased by 39.75%,which shows that the proposed design scheme can effectively improve the throughput and reduce the delay of the data center.