Research On Data Center Network Traffic Scheduling Based On OpenFlow

With the explosive growth of network traffic in data centers and the increasing proportion of horizontal traffic,traditional traffic scheduling mechanisms have gradually exposed the defects of insufficient link utilization and unknown network status.With the gradually emerging software-defined network architecture in recent years,it is easier to obtain the equipment and link status of the entire network,flexibly deploy traffic scheduling strategies,and then make fuller and more reasonable use of data center network resources.Therefore,designing efficient traffic scheduling algorithms based on software-defined networks is the focus of this research.This thesis studies the following two parts based on Fat-Tree topology,which is a common topology in data center network.(1)Firstly,for the scheduling of elephant flows in the data center network,a performance scale model is proposed.Meanwhile,Combined the advantages of self-organization of ant colony algorithm and fast convergence of simulated annealing algorithm,a data center traffic scheduling strategy based on ant colony simulated annealing algorithm was designed.By comprehensively considering the two factors of path hop count and path load,the algorithm actively avoids link congestion while reducing the proportion of idle links as much as possible.The simulation results show that the algorithm can fully utilize the link resources compared with the contrast algorithm,thereby more effectively improving the business performance of elephant flows.(2)Aiming at the scenario of comprehensive scheduling of elephant flows and mice flows in the data center network,a traffic scheduling optimization strategy based on alert flows mechanism is proposed.The alert elephant flows which are bandwidth-qualified are allowed to share the same link transmission with mice flows.This mechanism keeps the bandwidth utilization rate of the link occupied by the mice flows at a reasonable level while reducing the number of rerouting of the elephant flows as much as possible.Simulation results show that compared with the comparison algorithm,the mechanism can significantly improve the business performance of the mice flows,while sacrificing only a small portion of throughput of elephant flows.