| With the rapid development of the Internet industry and the rapid rise of related industries relying on the Internet,the TCP/IP architecture of traditional networks has an increasing difficulty adapting to the needs of network development.As a new type of network architecture,Software-Defined Networking(SDN)solves the problem of traditional network structure rigidity and management configuration difficulties by decoupling the data plane and control plane.With the continuous expansion of the network scale and the increasing number of streaming requests,the overload and underload problem of SDN has gradually become a key factor limiting the further expansion of its network scale.In the past,most researches on SDN load problems in the industry focused on the control plane,ignoring the same load imbalance problem in the control link.In the flow request process of the SDN network,the following three aspects of research still have shortcomings:(1)Before the flow request arrives at the SDN network,the prediction accuracy of the number of convection requests is insufficient,resulting in a problem that the control link that may be overloaded cannot be prevented in time;(2)The SDN digital control separation causes the OpenFlow Agent(OFA)module to not have high data processing capability,which in turn causes the performance control bottleneck of the uplink control link;(3)The Ternary Content Addressable Memory(TCAM)unit that stores the flow entry in the SDN is slow to update the flow table due to design reasons,which causes traffic congestion on the downlink control link.The optimization of the control link load is set as the research goal of this paper.The SDN network traffic prediction,uplink control link bottleneck and downlink control link congestion problem are analyzed and studied,and corresponding optimization and improvement measures are proposed to improve the bearer capability of the SDN system.The research work and main contributions of this paper are as follows:1.In order to solve the problem of uneven traffic prediction caused by uneven distribution of network data in SDN,an adaptive traffic prediction algorithm based on machine learning method is proposed.The algorithm combines the advantages of the Oboe system for adaptive adjustment of algorithm parameters and the Long-Short-Term Memory network(LSTM)algorithm for long-relevance data prediction.It can implement adaptive algorithm parameter adjustment for the network environment with changing time.Compared with the existing methods,the prediction accuracy is improved by at least 10%,and the adaptability of the algorithm is improved.2.An adaptive flow balancing algorithm is proposed for the Packet_In transmission bottleneck caused by the limited control link performance between the control plane and the data plane in SDN.The SDN network control plane has the characteristics of the global topology and the real-time state of the switch,and uses the threshold to control the start and stop conditions of the flow balance.By balancing the flow of each overloaded switch to the neighboring switch,only one hop is performed for each equalization.The adaptive method of forwarding solves the bottleneck problem of the uplink control link.Compared with the existing methods,the uplink control link load of 33% and the packet-in packet loss rate of 50% can be reduced,and the deployment overhead is small.3.Aiming at the problem of congestion of flow table caused by limited performance of data plane TCAM in SDN,an MPLS labeling algorithm based on flow redirection is proposed.The algorithm utilizes the source node switch flow redirection and the path switch to decentralize the MPLS label,occupies a part of the data link bandwidth to alleviate the problem that the TCAM update flow table rate is too slow,and can increase the system switch load capacity.Compared with the existing method,the link load is reduced by 20% and the system capacity is increased by 30% when the flow table entry insertion delay is constant. |
PDF Full Text Request