| With the continuous development of cloud computing,virtualization technology for integrating and managing physical resources is also developing rapidly.As a result,there are more and more application services in data center network,and the data flow generated increases rapidly.In order to meet the huge demand of data flow,data center network adopts Fat-Tree,a redundant topology with multiple forwarding paths.In order to realize the fair utilization of network resources in this topology,the algorithms often employed in the past are based on the traditional shortest path algorithm or equal cost multi-path algorithm(ECMP)to realize the scheduling of data flow.However,the issue of these algorithms is that the sudden increase of data flow will immediately occupy a large amount of resources of network equipment,result in problems such as link congestion or network equipment downtime,which may lead to immeasurable losses.Software defined network(SDN)is a new network architecture.It provides a centralized view of distributed network status for network traffic scheduling.Therefore,based on SDN,this thesis proposes two elephant traffic scheduling strategies for data center network in order to effectively control network congestion.The main works of this thesis are as follows:1.An elephant flow scheduling method DE-SDN is proposed based on differential evolution algorithm.The mathematical model is first built of the elephant flow scheduling problem.Real-time network information is then collected through s Flow technology,there by the SDN controller can quickly distinguish elephant flows from mouse flows.After that,according to the actual requirements of network traffic scheduling,the differential evolution algorithm is redefined to solve the mathematical model of traffic scheduling,so as to obtain the path of rerouted data flow.2.An elephant flow scheduling method DE-ACO is proposed which combines differential evolution algorithm and ant colony algorithm.In order to deal with the local optimal problem that the elephant flow scheduling algorithm based on differential evolution algorithm may occur due to serious individual assimilation in the later stage,this thesis firstly calls the differential evolution algorithm to obtain multiple rerouting paths,which are used as the initial pheromone of the ant colony algorithm.And then the DE-ACO carries on the second optimal screening,thus a global optimal path is obtained.The DE-ACO algorithm can effectively prevent the local optimal and secondary congestion problems in traffic scheduling by DE-SDN algorithm.3.Finally,the extensive experiments have been carried out to verify the effectiveness of the two elephant flow scheduling algorithms proposed in this thesis.In comparisons with several algorithms in terms of the average bisection bandwidth and the maximum link utilization,the proposed two algorithms can make the network load of data center network more balanced and reasonable,and improve the network bisection bandwidth. |
