| With the continuous expansion of communication networks and the growing demands for communication connection,traditional distributed control logic is difficult to cope with the increasingly complex network management requirements as well as optimizing network resource allocation.Centralized control is a network management manner relative to distributed control manner.The central controller brings more possibilities for network design and management based on its programmable capability.Accurate control of the network is achieved by implementing optimization strategy for the entire network.Not only that,different optimization goals can be obtained through flexible control over the network.In this thesis,we discusses the benefits of centralized control for network management and resource allocation in WAN(Wide Area Network),Campus network and more specific network structure in the Internet,and choose IP over WDM,Hybrid Software-defined Networking(HSDN)and datacenter networks as the representatives to do specific researches.This thesis focuses on the resource allocation issues while adopting centralized control over the networks.(1)Virtual topology design for IP over WDM using centralized controlAs the key component of WAN,IP over WDM mesh network can greatly improve the network transmission capacity.Previous studies regarding virtual topology design in IP over WDM mesh network always focus on energy efficiency,which provides no guarantee for comprehensive performances.Since centralized control can enable the consideration of different network performances simultaneously,it is easy to obtain optimal virtual topology for the two-layer network.In view of this,this chapter utilizes the centralized control over the two-layer network and proposes E~2VTD scheme to jointly optimize energy consumption efficiency,resource efficiency and network survivability of IP over WDM network.The simulation results show that compared with the previous algorithms,the three network performances are optimized by 39.8%,28.2%and 35.7%respectively.(2)Traffic Engineering and Manageability for Traffic in HSDNAs an emerging networking paradigm under centralized control,SDN is incre-mentally deployed into campus network,which is a preferable choice to make balance between increasingly complex network management requirements and the difficulties in full SDN deployment faced by network operators.The existing work only focuses on the improvement of traffic engineering(TE)performance,ignoring manageability of all the flows in HSDN.This thesis targets the both two problems,and utilizes some newly proposed technologies to realize centralized control over the entire HSDN,which is the premise for solving the two problems.In unicast communication aspect,a novel heuristic algorithm named Flow Routing and Splitting(FRS)is proposed.Benefiting from the advantages of centralized control,FRS makes all flows traverse at least one SDN node to realize central management over all flows,and enables HSDN to obtain TE performance comparable to SDN when deploying 20%SDN nodes.In multicast communication aspect,the explosive growth of multicast traffic together with the adoption of HSDN in campus network make resource allocation an open issue for investigation.Aiming at improving the TE and safety performance for HSDN multicast,Multicast Routing and Splitting(MRS)algorithm is proposed.For each multicast group,MRS calculates a suitable SDN node acting as an interface to central management.By deploying security policy through the interface,the safety performance can be enhanced.Then,MRS optimizes the traffic splitting ratios on all the multicast trees calculated per group to achieve load balance.Simulations show that,when deploying 40%SDN switches while calculating 2 trees per group,HSDN multicast adopting MRS can obtain a comparable TE performance to SDN multicast.(3)Managing Flow Transfers in Datacenter Networks based on centralized controlSince datacenter is the key infrastructure supporting big data and cloud computing,we choose datacenter networks as the specific network in the communication networks to explore the benefits of centralized control.When performing resource allocation in network layer,datacenter needs to consider the characteristics of the application in order to improve resource utilization and speed up application response.Accordingly,implementing centralized control over the application and network layers is the best choice.In view of this,Flow Chasing algorithm(FCA)is proposed in this thesis to optimize the data shuffle phase in datacenter cluster computing framework.A set of flows indicating the beginning or end of a calculation phase are defined as a transfer,and the goal is set to minimize the average transfer completion time(TCT).The evaluation shows that FCA can reduce TCT by 6.24%on average. |
