Research On TCP Congestion Control Mechanism And Bandwidth Allocation Algorithm In Data Center Networks

With the rapid development of cloud computing,cloud users can access resources such as computing,networking,storage,and software in a configurable resource pool on demand over the network anytime and anywhere.Data center network as the communications infrastructure is a network which interconnects large-scale servers in a data center for the transmission and exchange of large amounts of data between servers.Data center network is the core of the data center which has already become the focus of attention and research.Currently most of the data centers employ the scale-out model to interconnect the cloud computing servers,completing the tasks through the collaborative communication between the servers.Frequent communication between servers is required.Therefore,the performance of data center networks has become one of the important factors affecting the quality of service of cloud computing business.Cloud computing data center network has the features such as high bandwidth,low latency and under the ubiquitous many-to-one communication mode,Transmission Control Protocol may cause throughputs collapse.Meanwhile,the transmission performance requirements of the mixed traffic in the data center network are different.Mice flows are easily influenced by elephant flows missing the application deadline requirement.On the other hand,Traditional TCP which allocates bandwidth using flows cannot provide strict performance isolation for tenants in data center.The normal tenant application is susceptible to be interfered.In order to solve the above problems,it is necessary to study the TCP congestion control mechanism and bandwidth allocation algorithm of data center networks.Widely deployed applications in cloud computing with many to one communication pattern may suffer TCP Incast congestion problems resulting in throughput collapse.Mice flows are easily influenced by elephant flows missing the application deadline requirement.To address these problems,we propose the differentiated flow transmission control protocol.TCP flows are classified by DFTCP at the end server side,when network congestion happens,to reduce the high completion time of mice flows,DFTCP adjusts TCP congestion window based on network state and classified flow information.The simulation results show that DFTCP can effectively avoid TCP Incast congestion compared with the traditional TCP,and it can reduce the completion time of mice flows compared with DCTCP.For the multi-tenant data center network,we propose a minimum access guaranteed bandwidth allocation mechanism.To provide the minimum bandwidth guarantee and make full use of idle bandwidth for tenants,our scheme schedules VM's traffic through Weighted Fairness Queuing at the sending side,and it adjusts TCP flow's receiving window at the receiving side.The simulation results show that compared with the traditional bandwidth allocation mechanism which using flow,our scheme can achieve performance isolation,and compared with the static reservation allocation method,our scheme can improve the network utilization.The differentiated flow transmission control protocol and the minimum access guaranteed bandwidth allocation mechanism designed in this thesis can improve the network performance effectively and have some reference value for the research and application of data center network congestion control and bandwidth allocation.