| Over the past several decades,the Internet has made brilliant achievements with the help of TCP/IP protocol architecture,and has become an indispensable part of human lives.And the critical reasons of its success include IP-based packet switching technology and IP-based distributed routing control technology.On one hand,IP protocol not only adapts to the complexity of underlying protocols,but also provides support for the diversity of upper applications.On the other hand,distributed routing architecture makes edge access and regional interconnection with good scalability and growth.Meanwhile,its "distributed autonomy" feature makes the Internet obtain attributes of flexible survivability and fast self-healing,and it has become the underlying network architecture.However,distributed routing control framework also faces with some challenges including the ossification of calculation strategy of routing path,the lack of flexibility of flow scheduling,and the difficulty to dynamically adapt to the changes of network status.Especially,the growth speed of network traffic is far more than the replacement speed of network infrastructure.Therefore,distributed control architecture hardly makes full use of limited resources to achieve the optimization of network performance.Meanwhile,the separation between upper network services and underlying network resources results in less flexible customization capability of network service,and difficulty to adapt to changeable network environment.Fortunately,the centralized control architecture represented by SDN,completely separating control and forwarding plane,constructs network model including forwarding,control and orchestration,which can orchestrate network driven by network target to simplify the complicated configuration management tasks of the traditional network.But the centralized control architecture has potentially unavoidable problems,such as reliability,security,and routing convergence performance.Thus,this paper focuses on the coupling of distributed routing system with centralized control architecture in the routing control system and main contributions are as follows.First,we propose a software-defined routing system with integration of distribution control and centralization control,as called SDRS.The centralized control mechanism is introduced into routing control system based on distributed routing protocol,which is used to realize the coupling of closed-loop control of traditional network with the openness control of centralized control mechanism.On one hand,it achieves autonomous routing model by using closed-loop routing control in distributed routing protocol,and makes full use of its characteristics including flexible survivability and fast self-healing to establish reachable paths for end-to-end communication;on the other hand,it achieves cooperative routing model by using the centralized control architecture to perceive network status via distributed routing protocol and to calculate routing strategies that makes network performance tend toward optimized target,which fully utilizing the global network view and route orchestration capability in centralized control architecture to provide better quality of service.Finally,it uses inclusion principle and substitution principle to solve routing strategy consistency problem together generated by autonomous routing model and cooperative routing model.Thus,SDRS constructs a routing control model combining distributed control and centralized control,which making network control more efficient,flexible and accurate.Secondly,we propose a routing mechanism based on maximum multipath overlay tree by introducing multi-path routing into SDRS,as called SDRS-MMOT.Centralized controller calculates loop-free centralized path set by using maximum multipath overlay tree algorithm based on distributed paths generated distributed routing protocol.And it selectively distributes centralized paths into network according to orchestration strategy to achieve transmission of data-flow via distributed and centralized paths in parallel.In order to solve packet out-ordering problem,Hash bucket based path selection strategy is proposed to ensure that data-flow to the same destination is transmitted via the same path.At last,it uses distributed routing protocol to revoke centralized paths generated by centralized controller by extending traditional distributed routing protocol,to avoid the degradation of flexible survivability introduced by centralized control framework.Our experimental results prove that performance of SDRS-MMOT is superior to that of distributed routing protocol when network having congestion or single point of failure,e.g.if packet loss rate in distributed routing protocol is 20%,packet loss rate in SDRS-MMOT system is just 2.94%.Thirdly,we propose a lightweight online traffic engineering mechanism by introducing traffic engineer into SDRS,as called SRDR-LOTE.Centralized controller perceives real-time traffic distribution and calculates auxiliary path for congested links by using lowest cost flow-balance algorithm,while distributed routing protocol is responsible for calculation of main path.When network occurs congestion,the transmission of data-flow ships from main path to auxiliary path,to effectively solve link congestion problem.Meanwhile,SDRS-LOTE enhances IP head with good IP affinity to ensure that main path and auxiliary path don't interfere with each other,and uses a set of routing protocols to achieve dynamic traffic scheduling to avoid the complexity of maintenance of the two routing protocols in traditional traffic engineering.The revocation task of auxiliary path established by centralized controller is executed by distributed routing protocol,which is similar as SDRS-MMOT.Our experimental results prove that the performance of SDRS-LOTE is better than that of distributed routing protocol when network has congestion,e.g.when packet loss rate in distributed routing protocol is 48.7%,packet loss rate in SDRS-LOTE is just 2.2%Fourthly,in order to meet requirements of functional reconfiguration of data plane in SDRS system,a two-phase forwarding instance mapping mechanism is proposed by introducing flow switching plane into data plane on basis of network functions virtualization technology and modularization idea,which is used to deploy forwarding instance into equipment cluster.At initialization stage,static mapping mechanism calculate mapping relationship between logical function block and physical equipment based on resource mode of data plane,which also deploying these logical function blocks into corresponding equipment and determining interconnection relationship of logical function blocks by configuring service chain.At run phase,dynamical shipping mechanism can move deployment location of logical function block based on resource revaluation model,to optimize resource utilization.Our experimental results prove that best matching mapping algorithm is better than that of the other four mapping algorithms,and the range of shipping time of logical function block is [2.3s,55.6s]: logical function block without strategy just consumes 2.3s to complete shipping from one equipment to another equipment;and logical function block with strategy also uses 53.25 s to achieve strategy synchronization when the scale of routing tables is six hundred thousand.Finally,SDRS prototype system is built by using open source software including Click module router and Quagga routing suite,which is described from the aspects of functional plane and interaction protocol.The basic performance of SDRS prototype system includes: 1)packet forwarding ability with 7.1Gbps when forwarding 64 B packet with the help of DPDK optimization technology and start time with about 2.3 seconds;2)routing convergence performance with about 24 seconds when network diameter is eight;3)path maintenance time with 0.5 times of centralized control framework;4)network-status perceiving time with about 0.35 seconds when the number of nodes is 36. |
PDF Full Text Request