A Research Of The Scalability Problem In Software-Defined Networking

Software-Defined Networking(SDN)separates the control plane and data plane of the conventional network devices,leaving the network devices with only the data plane functions(e.g.,packet forwarding)while using a remote centralized controller to implement the control plane.This separation enables independent evolution of the data plane and the control plane,and facilitates the innovation of network functions.In SDN,the applications installed in the controller are responsible for controlling and managing the network.In this way,network events such as new flow arrivals or link failures can be processed flexibly,and it is possible to update control policies in the runtime.Meanwhile,the results of processing the events are expressed in the form of flow table entries which are deployed in the switches,e.g.,access denied or the routes.SDN supports the use of multiple protocol header fields to design the flow table rules,and thus enables fine-grained flow control.However,in order to provide the above abilities,SDN faces the scalability issues in its design.Specifically speaking,as the network scale increases,the controller will process more network events and thus may be overloaded,disabling the control plane functions.In the data plane,the use of more header fields will cost more storage space for each entry and there will be more entries in total.But the TCAM(Ternary Addressable Content Memory)in the switches has limited storage to hold the flow table.As a result,the whole flow table may not be stored in the data plane,causing that the control plane policies cannot be deployed in the data plane.The scalability issues in SDN limit the network scale that SDN can manage.If these issues are not addressed,SDN can hardly be used in large-scale practical networks.This paper focuses on the scalability problem in SDN and consists of the following works.1)After surveying the current SDN controller architectures and analyzing the limitations of the single-controller and the distributed-controller architectures,a novel controller architecture is proposed,with an event dispatching and scheduling algorithm based on the stochastic optimization model to balance the controller load.This works guarantees a scalable control plane while also being capable of balancing the load in time and at a fine granularity.2)For the data plane scalability issue,after analyzing the current approaches,a method on optimizing the timeout of the entries is developed.The basic idea is to evict the stale entries whose matched flows have finished,so that the TCAM space can be reused to hold more entries.The TCAM is modeled as a queuing system and the closed-form formulas are derived to quantitatively analyze the influence of a timeout value.Further,these formulas are used to find a proper timeout for the new flows.By this means,the TCAM storage is not wasted to hold stale entries and can be reused by more flows,alleviating the data plane scalability issue.3)With a deeper analysis on the timeout mechanism and according to the measurement of the traffic in a real-world network,it is revealed that a fixed timeout value for all the flows is not appropriate.Thus,an adaptive timeout management approach is proposed which considers the estimated duration of a flow,its flow type(packet arrival pattern),and the current TCAM utilization as a whole.With this approach,each flow obtains a proper timeout so that the stale entries do not occupy the limited TCAM storage and thus more flows can be supported by the same TCAM.By theoretical analyses and simulations,the effectiveness of the above works are verified.These works contribute to solve the scalability problem in SDN,and thus they will prompt the deployment of SDN in large-scale networks.