Research On Network Scheduling Used In Data Center

With the rapid development of the network,users have gradually increased use of the network,and various emerging applications such as cloud computing,big data,and virtualization have increasingly demanded the network.The requirements for the network are getting higher.Users hope that the network will bring a smoother network experience.The various enterprise-level applications need a stable and efficient network environment.In order to achieve various demands on the network,it is necessary to increase network throughput,reduce network delay,and try to avoid network congestion.Accurately apply for bandwidth allocation and routing settings for users.In order to meet the different needs of different users on the network,network traffic scheduling is particularly important.The internal traffic scheduling of the data center network realizes network load balancing,minimizes congestion,and reduces network delay.Traffic scheduling between data centers accurately captures the quality of user experience,and coordinates the traffic demand of each data center to actively implement forwarding strategies.Therefore,this article will study how to schedule traffic in the data center network to improve user experience,increase network throughput,reduce latency,and improve overall network performance.This article respectively focuses on the research on the traffic scheduling within the data center network and between the data center network.1.In the first part,we studied how to schedule traffic,average load,avoid congestion,and deal with network uncertainty in the data center.We should deal with such as dynamic flow and topological asymmetry and other issues.Data center network development trend marginalizes network functions.And because of the rapid development of advanced network card technology,load balancing based on edge deployment can be realized.This paper proposes a load balancing algorithm MEET,which is deployed on the end and based on crowd-sourced information.MEET is deployed at the end,can use the passive information of the network to collect the network situation.It can also conduct active detection,and at the same time fully perceive the actual network conditions at a low cost.Then according to the collected network congestion,the traffic is dynamically routed.After performing large-scale simulation verification,it can be obtained that the performance of MEET is 10% higher than that of Hermes,which is higher than that of DRILL in the case of a symmetrical topology.The average flow completion time has been reduced by 45%.2.In the second part,we studied how to schedule traffic between data centers.Modern wide area networks increasingly rely on software-defined networks to coordinate traffic transmission between various data centers.Software-defined networking can achieve lower latency,improve network utilization,and increase network throughput.In centralized traffic engineering,the above optimization goals can be achieved by adjusting the priority of the allocation of network bandwidth and routing settings.However,simple priority cannot accurately provide users with a better quality of experience.This article proposes QTE,through the coordination of various data centers traffic to implement the forwarding strategy.This paper first formulates the problem of maximizing user Qo E into two optimization procedures.Then proposing corresponding the heuristic algorithm which is implemented on the Floodlight controller system.And a lot of simulations have been carried out to show that QTE increases user experience 45.3% compared with previous work.