Research On Architecture Of Software Defined Data Center Networking And Key Technologies

With the gradual maturity of cloud computing technology and rapid deployment of emerging services such as big data and internet of things,the data center as their core supporting infrastructure grows rapidly.It is extremely urgent to build innovative data center to meet new business requirements of cloud computing and big data.Currently,cloud computing tasks usually need massive data flow interaction among commodity servers in data centers,and the data flow also shows some unique features compared with traditional network traffic,such as the round trip is pretty short,many-to-one or many-to-many communication exist extensively and so on.In such a situation,classical routing protocols with successful implementation experiences in traditional networks can't fulfill their roles in modern data center networks.Furthermore,Infrastructure as a Service(IaaS)as a core business in data center,it requires that the physical network resource should be dynamically scheduled.Therefore,the network resource such as address space,forwarding table space and link bandwidth should be virtualized.The service requirements and unique features mentioned above have brought great challenges to traditional data center networks.The Software-Defined Networking(SDN),as a rising network paradigm has laid the resolution of architectural needs in data center.It has innovatively separated the control plane from the data plane that makes networks more software-oriented and more agile.Furthermore,it provides virtual multi-tenancy,multipath forwarding mechanism,effective scheduling resources and centralized control strategies.However,the construction of data center network by applying SDN technology is facing severe challenges for the explosion of network information in data center.First the features of flow storage and lookup based on TCAM device have led to the restriction of scalability,energy consumption and high cost.Then the flow-based SDN mechanism triggers the control plane more frequently compared with traditional network,which has led to greater delay and overhead of bandwidth for the information communication between the controller and switches.As the limitation of the controller processing capacity and channel bandwidth,the network performance bottlenecks can easily occur.There are scalability problems such as high decision-making cost,bad controllability and reliability when applying the scheme in large-scale data center network environment.And further the structure of data center networking usually provides basic capacity guarantee,how to design efficient routing mechanism to improve network transmission efficiency and realize load balance ability is one of the key research directions.In this desseration,we focus on the following three problems:1)flow table storage capacity,2)control plane scalability,and 3)routing mechanism.The research achievements are summarized as follows:(1)A new type of data plane model named vector address based data center(VADC)is proposed which is based on vector exchanging technology.In VADC model,the VADC manager is to realize the logic function of the data center business for the whole multi-tenant.Meanwhile,it combines NOS and VM manager to unify storage resources and control calculations that can keep the synchronization schedule of network resources and calculation resources.There are also correspondence mappings from virtual machines for tenants to the network resource information that enables users control their own virtual network topology and make the management and scheduling by their own strategies.On the basis of the OpenFlow technology,the architecture employs an innovative source routing address named VA as the switching label of packets which defines the communication path independently.Thus,switches can accomplish the data forwarding process solely relying on VA.We use OpenFlow switches as gateways in network edges which are responsible for routing request and flow table lookup.As the data flows concentrating when comes to the core network,we take advantages of self-developed vector switches in high speed data forwarding process with which there is no need of table lookup and storage operation.The number of signaling consumed in flow table updating procedure can be greatly reduced.(2)A source-controlled data center(SCDC)network model which applys hierarchical multi-domain network management mechanism with multiple controllers is innovatively presented.There are one or more network domains in the network model which are connected with each other.The control plane is composed of one or more local and upper controllers.The upper logic links are generated from the topology aggregation of links among network domains.We reform the Ryu controller to achieve the controller functionality.Controllers extend part components of basic control functionality such as Ryu components management,OpenFlow protocol analysis,event dispatching and servers provided.Meanwhile,the features of VS management and VA collecting operation are newly added.Then on this basis,three components modules in SCDC control plane are designed which consists of topology management module,routing management module and northbound component module.We introduce network domain division and topology aggregation strategies to make the large-scale DCN into multiple miniature networks which makes it possible to reduce network decision costs greatly.In routing management module,a cost-amending multipath routing scheme is designed for intra-domain routing selection.On the basis of Dijkstra algorithm,multipath distance problems are solved by penalizing weights for selected paths according to the principle of isoline.The experimental results showed that the network architecture can well reduce network loads for control plane and improve network scalability significantly.(3)A node-disjoint multipath routing algorithm is proposed.The scheme combines the features of reverse AODV mechanism and On-Demand Node-Disjoint Multipath Routing scheme to determine available node-disjoint routing paths with minimum network routing control overhead.Meanwhile,it appends the backup routing tactics to make the procedure of data salvation more effective in case of link failure.Neighbor nodes adopt routing synchronization mechanism of recording,asking and listening,that has improved probabilities of being answered for RREQ messages and reduced the network overhead effectively.The experiment results obtained through various simulations show that the availability and superiority of the proposed strategy in terms of route availability,packet delivery ratio and control overhead.The proposed multipath routing strategy can well meet the requirements for building new data center networks which promotes the implementation and rapid development of new protocol,new business and new functions for future data center network.