Research On Load Balancing And Congestion Control Mechanism In SDN-based Data Center

With the rapid development of cloud computing,the traffic in data center grows exponentially.The load in the network is extremely imbalanced,resulting in network congestion.Due to increasing demand of users for quality of service,the studies of load balancing and congestion control are getting more and more attention.Nowadays,most load balancing strategies allocate the same number of flows on all paths,without considering flow size.Load is imbalanced as the traffic increases,resulting in higher latency and packet loss.This thesis mainly focuses on the study of the load balancing and congestion control mechanisms in dater center under software-defined networks.The research work on the load balancing strategies commonly used in data center network has been done.There are two types of the load balancing strategies: server load balancing and link load balancing.In order to solve the problem of flow mapping conflicts,the load balancing strategy based on flow classification(LBFC)is proposed.LBFC distinguishes between large and small flows by dynamic threshold.Different solutions are selected for different flows to solve the problem of high controller overhead caused by the fixed threshold.The schemes of fixed threshold are not suitable for the dynamic change of the data center traffic.At the same time,an adaptive routing algorithm based on probability is used to select the forwarding path for large flows,which reduces the possibility of assigning multiple large flows to the same path.Simulation results show that the LBFC can improve performance in terms of network throughput and transmission delay.When a host sends a request to a group of nodes,these nodes respond almost simultaneously,which results in a large number of bursts.This problem is called the TCP Incast.In order to solve this problem,two types of congestion detection mechanisms based on ECN(Explicit Congestion Notification)and RTT(Round-Trip Time)are studied.Existing ECN-based schemes is usually fixed ECN threshold which can only reflect the switch queue status.The RTT-based solutions need more accurate RTT measurements and the specific hardware support.Therefore,the congestion control scheme TDCTCP is proposed.TDCTCP determines the congestion degree according to the ECN and RTT.Different solutions are selected to adjust the size of sending window according to different congestion degree.TDCTCP can solve the problem of inaccurate congestion control signals and coexist with TCP flows.The simulation results show that RTT and average flow completion time can be reduced compared with other schemes.