Research On The Traffic Management Technologies Of Data Center Networking For Cloud Computing

The application of cloud computing and virtualization technologies make the "East-West" traffic of Data Center Networking for Cloud Computing increasing rapidly,and the network presents some new features in the aspects of virtualization,traffic pattern and topology structure.Because of the neglect of the internal characteristics of data center networking for cloud computing,traditional traffic management technologies usually cause the network congestion,transmission delay increase and throughput degradation,which affects the quality of service and user experience.Therefore,how to optimize and manage the internal traffic of data center networking for cloud computing has become the focus of researchers.The goal of the traffic management technologies of data center networking for cloud computing is to improve network performance and quality of service.Considering that the location of virtual machines in the network has a great influence on the distribution of traffic,as well as the difference in the length and transmission demand of different flows,the researchers have designed technologies,such as network aware virtual machine placement,multi-path routing and transmission control to manage the traffic of Data Center Networking for Cloud Computing.The above research has made great progress,but there are still many problems,such as high communication cost,uneven link load and multi service transmission performance.Supported by the National 863 project "Software Defined Networking Architecture and Key Technologies Research",this paper takes the new features of the data center networking for cloud computing as the starting point,studies from traffic distribution optimization and transmission optimization two perspectives to solve the shortingcomings of exising traffic management technologies — network aware virtual machine placement,multi-path routing and transmission control.The main work and contributions of this paper are as follows:1.A traffic distribution optimization algorithm based on virtual machine placement is proposed to solve the problem that the rapid growth of traffic in the Data Center Networking for Cloud Computing which leads to the congestion of the link and the high cost of the whole network communication.First,in order to minimize the maximum link utilization and communication costs,a multi-objective optimization model is set up.Then,based on the integer programming model,the differential evolution algorithm is improved,and the four indexes of communication cost,maximum link utilization,hardware resource constraint violation and link capacity constraint violation are considered.A ?-relaxation based multi-group elite selection strategy is proposed to select the optimal virtual machine placement result.Finally,the simulation results show that the scheme can reduce the total network communication cost by 17.5% and reduce the number of congestion links by 25% compared with the existing network aware virtual machine placement scheme.2.In view of the low throughput and unbalanced load of the cloud data center,a multipath scheduling mechanism based on SDN is proposed.First,the host realizes elephant flow detection by marking packets whose length are beyond the TCP cache threshold.Then,the controller periodically collects elephant flows' information to make path optimization to achieve load balancing.In order to reduce the traffic between the controller and the switch,an elephant flow selection algorithm is proposed to reduce the amount of control traffic by improving the particle swarm optimization algorithm.In order to select the global optimal scheduling path,a global optimal adaptive algorithm is designed to optimize the elephant flows' path,and to improve the transmission efficiency of the flow,the Segment Routing technology is adopted to reroute the elephant flows.Finally,simulation results show that compared with the existing multi-path flow scheduling mechanism,the mechanism can increase throughput by 9.3% while reducing controller communication overhead by 35%.3.A mix flow scheduling mechanism based on slack time and accumulation is proposed to meet the transmission requirements of the deadline flows and non-deadline flows in the data center networking for cloud computing.Firstly,the delay tolerance of deadline flows on non-deadline flows is measured with the introduction of Slack Time.Then the queueing delay of non-deadline flows is reduced by completing deadline flows barely before their deadlines.Lastly,non-deadline flow is scheduled according to the number of bytes it has sent to reduce the average flow completion time.Experiment results show that the proposed mechanism can effectively reduce the average flow completion time of non-deadline flows while maintaining low deadline miss rate.