Research On Traffic Identification And Scheduling In Data Center

With the deployment of cloud computing,virtualization and other emerging businesses in the data center,the internal ecology of data center is undergoing significant changes,which brings many problems.Emerging businesses require a lot of data communication between servers,resulting in a rapid increase in internal traffic.The traditional TCP/IP data center architecture is difficult to adapt to the new traffic scenarios,which causes frequent network congestion and packet loss and seriously affect service quality and service performance.How to design an effective scheme to manage a large amount of traffic in the data center,improve the network performance and reduce the cost of network management has become the research direction of this paper.As an innovative network architecture,SDN realizes the decoupling of logical control and physical forwarding.Compared with TCP/IP architecture,it has the global view of the network and the ability of fine-grained traffic scheduling,and is widely used in data centers.Optical switching technology has the advantages of high transmission rate,large throughput and low energy consumption,which can make up for the shortcomings of traditional electrical switching network.In this paper,SDN technology is used to solve the traffic problem in the data center from two aspects of traffic identification and scheduling policy.Firstly,this paper analyses the distribution regularity and type of flow within the data center.The data center network congestion problem is mainly caused by elephant flows which carry a large amount of data and last long.This paper proposes a traffic Detection scheme HCFD(host-controller Flow Detection)combining Host and Controller terminals to efficiently and quickly identify elephant flows.on the host side,HCFD use the Linux kernel Netfilter framework to mark the flows that exceed the threshold with ToS field tags.Then the network switcher will send the tagging flows to the SDN controller.On the controller side,HCFD use the decision tree classification model to make the secondary identification to the flow.Secondly,based on the architecture of photoelectric hybrid data center,this paper proposes a set of traffic scheduling scheme and algorithm implementation for two scenarios:Inter-ToR and Intra-ToR.To realize the traffic adaptation and switching mechanism of deep fusion,the optical link is used to transport the identified elephant flow which can ensure the throughput requirements of the elephant stream,while the electrical link is used to forward the mouse stream to ensure the low latency requirements of some mouse stream.Finally,relevant simulation experiments are designed to verify the feasibility of the elephant flow recognition scheme.Compared with the ECMP scheduling scheme with the traditional fat tree topology,experiment results show that the traffic scheduling scheme under the photoelectric hybrid architecture can effectively alleviate network congestion,reduce end-to-end delay of data and reduce packet loss rate.