Research On Elephant Flow Detection And Scheduling Strategy In SDN-based Data Center Network

With the advent of the cloud era,the business volume of data center network has been expanding,which has caused a great burden on network management and operation.People's lives are increasingly dependent on the network,and the demand for network resources is increasing.In the data center network,the problem of network link congestion and load imbalance caused by elephant flow is becoming more and more serious.Elephant flow is characterized by large load and large amount of data traffic in a period of time.For traffic scheduling,the main problem of the traditional Equal-Cost Multi-Path(ECMP)scheduling algorithm is that in the process of elephant traffic forwarding,a large number of data packets are easily congested to a link node,resulting in transmission link collision,network congestion,transmission delay and other issues.The emergence of SDN(software-defined networking)new network architecture has the characteristics of global network view and programmability,which brings new opportunities to solve these problems.In the SDN environment,the elephant flow will be detected first in the data center network,and then a new elephant flow scheduling strategy will be formulated.The main work of this paper includes:(1)A new ADU(Auto Detect Upload)elephant flow detection method is proposed.ADU detection mechanism is divided into two parts: ADU-Server and ADU-Client.ADU-Client runs in the host computer,mainly to monitor the host's network data transmission buffer queue,while ADU-Server runs in the controller as a module.When a host is about to send an elephant stream,ADU-Client generates a packet with forged source IP address and sends the information of the elephant stream to the controller by triggering the Packet_in message of the edge switch.The process of ADU's reporting depends on the Packet_in mechanism of Openflow protocol.The switch can send information directly to the controller through the Packet_in message,which achieves ultra-low consumption and ultra-high level.Efficient elephant flow detection mechanism.(2)An improved simulated annealing algorithm,a SA-GFF(Simulated Annealing-Global First Fit Algorithms)scheduling algorithm with better global optimization effect,is proposed.Firstly,the information of elephant flow is obtained by efficient ADU elephant flow detection method.First,the initial adaptation algorithm is used to initialize and schedule the new elephant flow.At the same time,the SA-GFF algorithm is started to optimize all elephant flows in the current network based on the current global link state.In this paper,a series of simulation experiments are carried out to verify the ADU elephant flow detection method and SA-GFF traffic scheduling algorithm proposed in this paper.The preliminary results show that ADU can detect elephant flow efficiently and greatly reduce the time cost.SA-GFF scheduling algorithm is superior to traditional algorithms and other heuristic intelligent algorithms,and can effectively reduce the maximum link utilization of the network and achieve links.Load balancing,network enhancement and so on.