Research On Generation And Update Of Routing Rules In Software Defined Networking

The expansion of network scale and the emergence of new network applications,not only has brought heavy traffic burden to the network,but also diversified the quality of service requirements.In recent years,although operators continue to broaden link bandwidth and speed up their ports,the lack of effective traffic management mechanism leads to very low resource utilization.Since the QoS(quality of service)requirements vary greatly between different flows,how to provide adaptive,differentiated and customizable routing for flows according to the real-time network states and the flow QoS requirements is the key to improve the network resource utilization and the QoS.Because of the tight coupling between the control plane and the forwarding plane along with the rigid architecture,it is difficult for the traditional network to adopt flexible strategies to meet the increasingly diverse flow requirements.SDN(Software Defined Networking)is one of the possible paradigms for future networks due to its separation of control,forwarding planes and programmable architecture,which injects great vitality into the evolution of network innovation.Moreover,the software defined networking can offer global network view and fine-grained control capabilities,which greatly reduces the difficulty of implementing adaptive,differentiated and customizable routing for flows.Relying on the National Key Basic Research and Development Program of China(973 Program)"Network Component Model and Clustering Mechanism",this paper focuses on the routing rules generation and update process in SDN.Based on the global network view and QoS-aware flow classification,the deep reinforcement learning technique DDPG(Deep Deterministic Policy Gradient)and the service-oriented path customization technique segment routing are introduced into the routing rule generation and update process to realize adaptive,differentiated and customizable rule generation and update.Furthermore,aiming at rule update process in the actual situation of SDN deployment,the operator will maintain hybrid SDN networks for a long time,a feasible and lightweight update algorithm for hybrid SDN is designed.The contributions of this paper are mainly concentrated on the following aspects:1.In order to generate adaptive,differentiated and customizable SDN routing rules,a DDPG-based SDN routing rules generation mechanism is proposed.With the fine-grained flow QoS classification and network state awareness,a feasible architecture for SDN to load machine learning is proposed.On this basis,DDPG machine learning technique is introduced into the routing rule generation process of SDN and a routing rules generation mechanism is designed.First,the routing rule generation mechanism formulates the optimal control strategy based on the QoS requirement of flows and network status.Then,the DDPG agent can achieve the black-box optimization in continuous time to generate the optimized routing rules according to the control strategy.At last,the performance of the routing rule generation mechanism is evaluated through experiments.When the strategy is to optimize the network delay,the network delay achieved by the proposed mechanism is reduced by 63.82 ms over the traditional OSPF protocol.When the strategy is to optimize the throughput,the proposed mechanism can increase the throughput by 12.5% compared with the existing technique.2.To implement adaptive,differentiated and customizable update,an adaptive update technique in SDN environment is proposed.Based on the analysis of the existing update mechanism and the consistency constraints of the update process,an update technique with multiple working modes is proposed.This technique abstracts the common update mechanism into corresponding update operations and calculates the update operation sequence at the node granularity.The emulation outcomes show that the performance of different work modes is different,and they are optimized in TCAM reduction,bandwidth consumption and update delay respectively.For example,when the Tri-fusion algorithm work mode is leveraged,it can reduce at least 85% TCAM overhead,improve 9%,65%,82% over other work modes and the comparing algorithm.Therefore,the network can adjust the working mode of the update mechanism adaptively according to the QoS requirements and real-time status,so as to adjust the consistency model,delay,TCAM overhead reduction and other performance parameters of the update process,optimize the key network parameter that restricts the performance improvement,and ultimately improve the network resources utilization and quality of service.3.To solve the problem of hybrid SDN update and promote the actual deployment of SDN,a low-overhead consistency updating technique for hybrid SDN is proposed.Based on the deep study of the routing properties of hybrid SDN networks,the segment routing is imposed into the update of hybrid SDN networks,and the application of segment routing in hybrid SDN network update is formalized.Combining the basic principles of two-phase commit and ordered scheduling,a general updating method for hybrid SDN networks is designed.This method can greatly reduce the TCAM overhead and time cost compared with the existing update techniques.For example,in the case where the distance vector routing is involving,the algorithm can reduce at least 99% time cost than the comparing algorithm,and make at least a 10% improvement over the comparing algorithm in TCAM overhead.