Research On Congestion Control Algorithms Based On Software Defined Network

With the rapid development of computer technology,applications put forward higher requirements for network transmission quality.However,traditional network transport protocols are throughput-oriented and thus are unable to meet the high-quality transmission requirements of these applications,especially those real-time applications.Specifically,there are a lot of types of flows in the network,the characteristics and transmission requirement of which are different.It is hard for traditional congestion control algorithms to have an accurate view of the underlying network,resulting in high delay and poor network utilization.The advent of Software Defined Network(SDN)provides a new idea for network congestion control with its features of network abstraction,programmability,central control.SDN-based hybrid flow scheduling algorithms can help to solve the problems of traffic scheduling and congestion control.How to apply SDN to dynamically schedule the network flow,provide feedback to end-hosts,and modify the transport layer protocol parameters according to different network flows is an important problem to be solved.The research work on the congestion control based on Software Defined Network has been done.The contributions of our work is summarized as follows:Firstly,we propose Software Defined Congestion Control Plane(SDCCP).Since end-hosts have limited knowledge of the underlying network,end-to-end congestion control algorithms usually fail to achieve high performance.In SDCCP,based on the global knowledge of the network in SDN,end-hosts can have a clear view of the network condition so that they can change transport layer behavior based on the network condition,which is possible to achieve better network performance.Secondly,we propose a Feedback-based Congestion Avoidance(FCA)algorithm which runs atop SDCCP.The feedback packet sent from the SDCCP controller can help end-hosts improve their network awareness.Based on the feedback end-hosts classify the network condition,and adjust their transport layer parameters such as congestion window and so on.This algorithm can efficiently reduce the queue length of the network,achieving high network utilization while keep the network delay in a low level.Finally,we proposed priority FCA(p FCA).Priority is assigned to flows based on the flow characteristic.Then flows with different priority have different transport control logic.By priority-based preemption,p FCA can guarantee the network performance of high-priority flows.However,there are still some problems such as heavy overhead,inflexible priority allocation,and low network link utilization under some workloads.The future research includes reducing extra overhead,dynamically assigning priority,and optimizing network flow scheduling algorithm.