Research On Elephant Flow Detection And Scheduling Based On OpenFlow Network

With the continuous expansion of computer networks and the ceaseless enrichment of application types,the network's structure together with the function is becoming increasingly complex.In order to solve the emerging network troubles,a new type of network technology named SDN generates.SDN is a new method to regulate programmable switch.In the data center network,with the help of central soft controller,SDN is able to achieve the overall network flow optimization and QoS management,etc.The elephant flows usually appear in the data center network and occupies mass bandwidth,so the detection and scheduling of elephant flows benefits the easing of network congestion.Taking advantage of SDN's centralized deployment and programmable performance,the OpenFlow network based elephant flow detection and scheduling makes it easy to load applications with real-time performance.This article elaborates researches on aspects of traffic detection and scheduling in the data center,by using SDN centralized control function,finally we designed the specific detection and scheduling mechanism to improve the network performance.(1)Based on the analysis of network traffic from data center,a two-stage real-time detection method is proposed to detect the elephant flows.Phase1,filtering the sampling packet in the switch kernel space.Phase2,obtaining dynamic-threshold identificating data flow according to Bayesian method.OpenFlow protocol and sFlow sampling regular are used to monitor network status and obtain link traffic.(2)In view of the detected elephant flow scheduling.According to the data center network topology in the multi-path common situation,in OpenFlow network,considering link bandwidth and shortest path routing,the proposed path selection mechanism of the elephant flows guarantees to make full use of network resource and design flow table rules for the elephant flow scheduling for high load link and it will ease the network congestion.(3)The prototype system is designed and constructed for elephant flow detection and scheduling that applies to the data center.Using simulation system verifies the feasibility and effectiveness of the elephant flow detection and scheduling method based on OpenFlow network.Combining Ryu controller,in the Mininet experiment platform,constructing data center based on OpenFlow network for experiment.In the conclusion,the results shows that the proposed elephant flow detection method based on OpenFlow network in this paper can quickly detect the elephant flow under different conditions.Based on analysis of link status after congestion scheduling,it shows that the method proposed for data transforming in this article is better than the traditional shortest path route based ECMP method.