Research On Sdn Migration And Optimization In SDN Transition

The transition from the traditional network to the Software Defined Net-working(SDN)has become a trend.The device substitution,the controller placement,the resource management and upgraded technologies for transition are key parts of the SDN transition.SDN provides fine grained traffic control,which can improve the flexibility of network control.By practical applications,such as Google’s B4,the performance of SDN has been recognized.However,due to the operational and economic constraints,the transition from traditional network to SDN cannot be completed once.The way of grad-ual migration is usually adopted for the SDN transition.It has an impact on the forwarding layer,the control layer and the application layer of the network.Different migration strategies bring different benefits to the network,thus,dur-ing the gradual migration,the optimization problems should be considered from different layers.For the forwarding layer,during the gradual transition,the net-work is in the state of a hybrid SDN,which contains legacy devices and SDN devices.The network is in a different state with a different migration sequence,which makes different benefits.For the control layer,the placement of con-trollers in a hybrid SDN has an important influence in the efficiency of the interaction between controllers and SDN forwarding devices.For the applica-tion layer,the allocation of resources in hybrid SDNs affects the performance of the network,such as the link utilization,the network utility and the Quality of Experience(QoE).Therefore,the SDN transition needs to be studied from the forwarding layer,the control layer and the application layer,respectively.In this paper,first,the migration strategy is studied from the perspective of the forwarding layer.Then,the controller placement problem in the hybrid SDN is studied from the point of view of the control layer.Finally,the optimization problems of network resource allocation are studied from the perspective of application layer.The main contributions and achievements are summarized as the follows.(1)Migration strategy from the traditional network to the SDN.In the gradual transition of SDNs,the devices migration sequence is sched-uled from the perspective of customers.The definition of customer’s ben-efit brought by migration is given,which considered multiple parameters including the number of new alternative paths,their quantity and the load of traffic from customers.The proposed method can give a good migra-tion sequence based on customer’s benefit.The experimental results show that the proposed scheme can maximize the customer’s benefit in each mi-gration stage.(2)Controller placement in hybrid SDNs.In a hybrid SDN,the placement of controllers affects the efficiency of network management.Control delay and load balancing of controllers are two important indicators for the controller placement.In this paper,where to put controllers and how to assign them to SDN forwarding devices are two significant issues.For the former one,controllers are placed with the minimum communication delay.For the latter,the SDN controller as-signment problem is formulated as a minimum weight matching problem of the bipartite graph,which considers both the delay and the load bal-ancing of controllers.The experimental results show that the proposed method can ensure both the minimum control delay and the load balanc-ing of controllers.Thus,control messages can be transmitted between the SDN controllers and SDN forwarding devices efficiently.(3)Resource allocation based on the optimization in hybrid SDNs.The hybrid SDN has the characteristics of multipath forwarding.In order to improve the network utility,the resource allocation in the hybrid SDN is studied.First,the resource allocation optimization scheme in the core net-work is proposed.This scheme can maximize network utility during the SDN transition,and it can improve customers’ satisfaction with the opti-mal bandwidth and path allocation strategy.Then,the resource allocation optimization scheme in the SDN 5G network is proposed.When the access network is based on the NOMA of 5G,in addition to resource allocation in the hybrid SDN core network,the sub channel allocation and power al-location in the access network should also be considered.The end-to-end network resource allocation aiming at maximizing network utility is pro-posed in this paper.By maximizing the network utility of hybrid SDN 5G networks,the proposed method can improve customers’ satisfaction with the allocated resource,which includes the allocation of paths,bandwidth,channels and power.(4)DRL-based resource allocation in hybrid SDNs.The resource allocation algorithm based on optimization is complex.Thus,it is difficult to adapt to the changes of the network quickly,especially,when the network is in a large scale.To deal with it,Deep Reinforcement Learning(DRL)is introduced into the resource allocation problem.It can make decisions quickly based on the real-time interaction and its experi-ence.For multimedia services,a DRL-based resource allocation scheme is proposed in this paper.By the interaction between the agent and the environment,it can make intelligent schemes to maximize QoE.Mean-while,in order to obtain the feedback of customers’satisfaction quickly,a prediction model of QoE is proposed.This model can quickly provide the feedback for the adjustment of resource allocation strategies.Thus,the performance of learning can be greatly improved.The experimental results show that the proposed scheme can greatly improve QoE of multi-media services.