Traffic Scheduling For Public Cloud Based On Flow Prediction

With the explosive growth of social information,the maturity of resource reuse technology and the popularity of broadband networks,the birth and development of cloud computing has been promoted.As an important part of cloud computing,cloud data center network is a bridge connecting all servers in the data center,and its performance directly affects the quality of cloud services.At the same time,large-scale cloud computing is generally accompanied by massive data transmission,which brings many challenges to the infrastructure cloud data center network supporting cloud computing.However,due to the multiple services running in the existing data center and the multi-tenant working environment,many traditional traffic scheduling methods have encountered many problems when running in the data center.For example,traffic in a multi-tenant data center is sensitive to delay.At the same time,according to the shortest job priority algorithm,we should prioritize delay-sensitive traffic and traffic with a small flow length;in addition,due to the host system in the cloud data center or hardware in the network,users may adopt different network protocol stacks,which will cause some users'traffic to be at a disadvantage in the competition of network resources.This paper has made research on the data center traffic scheduling problem and achieved the following research results:1.An Information-.agnostic Flow Scheduling(IFS)scheme is designed.It is hard to timely identify latency-sensitive flows for information-agnostic flow scheduling.Thus,we design IFS scheme to identify flows and decrease the flow completion of latency-sensitive flows.We use DBSCAN as our cluster algorithm,and use flow information such as IP address,TCP ports,to classify the flows in public cloud,then transmits the flow priority policy to all switches by a centralized master node.Finally,the testbed evaluations show that IFS can reduce the completion times for latency-sensitive applications.2.A TCP flow mapping scheme to improve TCP's performance is designed.Data Center TCP(DCTCP)is specifically designed for Data Centers,which makes it have better performance than traditional TCP.However,DCTCP is not supported in all systems and devices.For some old operation system,it's hard for user to update their system and applications.We use virtual switch ovs to make TCP flow can support ECN as DCTCP,meanwhile,we do not change any system or applications of users.Our testbed evaluations show that we improve TCP's performance,and enhance the fairness of the network.