SDN Performance Optimization Based On Flow Entry Timeout

The expansion of network scale and the rise of smart phones increase the complexity of networks and the difficulty of network management.Traditional network devices couple the control and forwarding functions,it is difficult for administrators to maintain the global network view,and based on which to enable network performance and resource management in a global manner.Software Defined Networking(SDN)addresses it by providing a layered network architecture,decoupling network control and forwarding functions,centralizing network control,standardizing management interfaces,and maintaining a global view of the network.In SDN,forwarding devices(SDN switches)in the data plane have no control capabilities,they rely on flow entries generated by controllers in the control plane to complete the forwarding of the data packets.The installation of flow entries consumes the control channel bandwidth,and resources of switches.Some flows such as elephant flows consume most of the network bandwidth.Multiple elephant flows sharing a same link may cause network congestion and high link bandwidth consumption.Therefore,the optimization of control channel bandwidth usage,the number of active flow entries in switches,and the detection accuracy of elephant flows have become the research hotspots in SDN.Currently proposed optimization solutions typically have one or two optimization objectives without fully considering the tradeoffs among multiple objectives.For instance,detecting elephant flows requires controllers to poll flow statistics from switches,however,moving flow statistics from switches to controllers consumes a big amount of control channel bandwidth,leaving less bandwidth to other control messages and leading to an increased network forwarding delay.To manage network forwarding delay,a longer timeout may be given to flow entries.However,a longer timeout implies that flow entries have longer life span in switches.It may waste the resource of switches,overflow the flow table of switches,and lead to a significantly increased network delay.It is significant to jointly optimization multiple objectives to optimize the network performance and resource usage.This work uses the packets forwarded by switches to controllers for the installation of flow entries to directly detect the elephant flows.By adjusting the type and value of timeout of flow entries,this work jointly optimizes the bandwidth usage of control channel,the number of active flow entries at switches,and the detection accuracy of elephant flows for the management of network performance and resource usage.Particularly,this work firstly constructs an elephant flow model,then formulates the three optimization objectives.An novel timeout type that hybrids hard and idle timeouts and two timeout adjustment methods,the mean variance and the double methods,are given based on the flow entry timeout mechanism over SDN networks.The optimization problem is solved using NSGA-? and Bayesian multi-objective optimization algorithms.Extensive evaluations based on the real packet trace captured in Guilin University of Aerospace Technology are provided to estimate the effectiveness of the timeout method and the adjustment scheme.Although the proposed hybrid timeout can manage the network performance and resource usage better than the hard and idle timeouts,the hybrid timeouts that adjusted using the doubling method achieves the best performance and resource management over SDN systems.Although both NSGA-? and Bayesian multi-objective optimization algorithms can efficiently generate approximation solutions for the proposed optimization problems,mixing the solutions of both algorithms can generate a approximation Pareto Frontiers with higher quality.