Research On SDN Load Balancing Based On Two Level Identification Of Elephant Flows

With the rising of cloud computing and big data services,the size of data center network and the number of users are growing rapidly,and the network flows present an explosive growth.But the traditional data center network based on TCP/IP architecture can’t achieve the efficient management of massive data.The problem of network congestion and the waste of network resources become more and more severe.How to achieve the load balance of the data center network to improve the transmission efficiency has become an urgent problem which should be solved.In recent years,the Software Defined Networking(SDN)technology,which is characterized by the separation of control and forwarding plane,has received extensive attention from the IT community by its powerful traffic control capability.The logically centralized SDN controller can achieve the efficient and fine-grained network traffic scheduling based on the global view of network,which makes SDN has a larger advantage zhan the traditional network based on TCP/IP architecture.In this paper,we study the problem of unbalanced load of SDN data center from the point of view of elephant flow identification.Firstly,aiming at the problem of high cost of the existing elephant flow identification method,a two-level method for elephant flow identification is proposed.This method presents a suspicious elephant flow detection algorithm based on TCP write-queue(Suspicious Elephant Detection based on Write Queue,SED-WQ)in the first stage,which identifies the suspicious elephant flow by monitoring the data size in the write queue of host and reduces the processing cost of the controller;And presents a real elephant flow detection algorithm based on duration time(Real Elephant Detection based on Duration Time,RED-DT)in the second stage,which identifies the real elephant flow by monitoring the duration time of the suspicious elephant flow in the network and improves the accuracy of the elephant flow identification by eliminating the unfit elephant flows.Secondly,aiming at the problem of load imbalance in SDN data center,a load balancing method for SDN network based on two-level identification of elephant flow ELB(Elephant Load Balancing)is proposed.Aiming at the elephant flow in the network,ELB presents an elephant flow scheduling algorithm based on uniform distribution,and plans the optimal forwarding path of elephant flow dynamically to ensure the efficiency of ELB.Aiming at the mice flow in the network,ELB presents an mice flow scheduling algorithm based on random forwarding,and plans the optimal forwarding path of mice flow statically to reduce the cost of controller.Finally,we use the Mininet software to simulate and analyze the two-level method for elephant flow identification and ELB.Analysis and simulation prove that,under the premise of ensuring the accuracy of elephant flow identification,the two-level method for elephant flow identification reduces about 85% cost of identification compared with sampling identification method.Meanwhile,under the premise of ensuring low identification cost,ELB reduces the average transmission delay by about 10% and increases the average utilization of links by about 5% compared with the traditional network load balancing method based on ECMP(Equal Cost Multipath Routing).