Research On Load Balancing Based Onprogrammable Data Plane

In recent years,the continuous introduction of "Internet +" thinking and the development of cloud computing technologies has resulted in an increasing amount of data being transmitted through the network.This has put greater demands on the performance of the network,and is reflected through indicators like delay,throughput,link utilization,network congestion control and load balancing capabilities.Traditional network equipment has limited flexibility,is not easy to expand and is expensive and time consuming to replace.The emergence of SDN(Software Defined Network),a new generation of network technology,brings new opportunities for the network.It separates the data plane and the control plane.The flow table rules in the data plane come from the SDN controller and are delivered through Openflow protocol.Additionally,SDN provides a flexible programmable interface which improves the ease of management and the performance of the network.However,there may be some delay in handling the message,including the processing delay in the controller and the forwarding of the flow table.When there is a surge of network users,the delay of message processing will lead to path congestion and affect the quality of network service.This thesis presents a scheme for programmable data plane load balancing based on path selection,which aims to break the limitations of the traditional network switches and make the data plane expansible.The thesis has used P4(Programming Protocol-Independent Packet Processors)language to program and optimizes the data plane processing with the aim of optimizing packet forwarding and load balancing.The experimental results obtained by using NS2(Network Simulator-version 2)show that the scheme is effective.The main contribution in this thesis is as follows:(1)A load balancing scheme based on data plane data flow was proposed.Aiming at the problem of uneven path load distribution,this paper proposes an improved ECMPalgorithm based on the traditional ECMP(Equal-Cost Multipath routing)algorithm.At first,the traditional ECMP algorithm is used to calculate the hash value of the path and flow ID,and then combine the current path bandwidth weight information to select the path.It avoids the limitation of being assigned to the same path under multi-path data flow,and improves the overall performance of the network.(2)The design and implementation of programmable data plane.Firstly,the design architecture of programmable data plane was introduced.Secondly,the challenges facing the current programmable data plane were analyzed.Finally,the processing logic of data stream forwarding is designed by P4 advanced language,and the programmable processing on the data plane switch has been realized,and completed the path efficient allocation and network load balancing scheduling.(3)NS2 simulation and result analysis.The NS2 simulation platform is used to simulate the proposed scheme in this thesis.The simulation results show that when compared with the traditional ECMP,the improved ECMP algorithm allocates the paths through the path bandwidth weight information to achieve the load balancing.Under the same network load,the average flow completion time of the improved ECMP algorithm is lower than that of the traditional ECMP algorithm.It can effectively save average flow completion time,reduce the end-to-end delay,and reduce the packet loss rate of data traffic,and improve the throughput of the network.In addition,the load balancing of the network has been realized.