Research On Large-scale Traffic Scheduling Technology Based On SDN

With the rapid development of new computing models such as Internet applications and cloud computing,big data,and artificial intelligence,the size and the amount of data flow of data centers,which are the core of modern information service infrastructure,are exponentially increasing,and the types of network services they provide are also complex.Thus,the network traffic management under the new architecture are facing two major problems.On the one hand,the multipath characteristics between the data center network nodes and the locally dynamic data flow make the traditional ECMPbased traffic scheduling mechanism easily lead to unbalanced link load,and excessive network conflicts may lead to poor network performance.On the other hand,due to the high cost,low scalability,and low flexibility of the traditional load balancing mechanism,its application on the underlying server cluster of the data center cannot achieve good distribution of traffic requests.Based on these two aspects,in order to improve the overall service quality of the network,this thesis aims to improve the overall service quality of the network by utilizing the full network view and the open interface for centralized control of SDN.The core research object is traffic scheduling of data layer in data center network based on software-defined network: path traffic load balancing and service request traffic load balancing,which mainly includes the following three aspects::(1)Aiming at the problem of path traffic load balancing,a traffic scheduling mechanism based on hybrid genetic simulated annealing algorithm is proposed.Firstly,the detection mechanism at the end server is introduced to detect and mark the large stream.It is proposed to use the bandwidth limitation of the end server NIC to design a large-flow real bandwidth demand prediction algorithm.The large-flow scheduling problem is reduced to the multi-clip problem,the improved hybrid genetic simulated annealing can be utilized.The algorithm searches for the global optimal path for the large stream.Finally,the OpenFlow switch uses the segmentation route to transmit the large stream of rerouting.The simulation results show that the mechanism can obtain better scheduling effect in the network environment with more currents and more frequent traffic conflicts.At the same time the average bandwidth,delay and throughput of the network are better.(2)For the service request traffic load balancing problem,a dynamic load balancing mechanism SBDLB based on SDN is proposed.On the one hand,the SBDLB mechanism can match the incoming traffic request with the flow table rules and can pre-match the traffic requests generated by the client in the future,which reduces the number of flow tables,the load on the controller and the load on the switch.On the other hand,The SBDLB mechanism classifies the incoming traffic requests according to the service type,collects the server cluster load information by using the sFlow protocol,and finally designs a load balancing algorithm with dynamic weights for load distribution.The simulation results show that the mechanism is beneficial to increase network throughput,to reduce the average response time,to speed up the processing of traffic requests,while the load between servers is more balanced,resource utilization is higher.At last,the purpose of optimizing network service quality is achieved.(3)Finally,the prototype system design is carried out for the above proposed traffic scheduling problem solution.The two mechanisms are combined,and the module is realized on the open source SDN controller Floodlight v1.2.The visual scheduling analysis system is written with the Flask web framework.A data center network traffic scheduling prototype system is implemented based on SDN.Combined with the Mininet simulation platform,it proves that it can obtain better network performance indicators and improve network service quality.