Research On Load Balancing And Energy Saving Of Data Center Network Based On SDN

With the large-scale deployment of data centers,the data traffic in data centers has continued to increase.Different types of traffic have different link requirements and service quality requirements,and most existing routing algorithms do not comprehensively consider the real-time status of link and various traffic characteristics,result in the overload of part of the links in the network while the other part of the links is still idle,which causing the imbalance of the network load.Compared with traditional networks,Software Defined Network(SDN)is a dynamic,manageable,cost-effective and adaptable emerging architecture.Its separation of transmission and control and centralized control make it possible that adjusting network management strategies with real-time network resource information.The network management strategy realizes the flexible control of the entire network.This provides a new idea for solving the load balancing problem of the data center.This thesis mainly studies SDN-based data center network load balancing and energy efficient issues.Firstly,in view of the problem that traditional multi-path routing algorithms are likely to cause conflicts of elephant flow with the result of network congestion,a multi-path dynamic load balancing mechanism(MTDLB)based on the SDN is proposed.MTDLB divides flows into elephant flows and mice flows,different routing methods are used for different kind of flows to meet the high throughput requirements of elephant flows and the low latency requirements of mice flows.At the same time,a random path method based on path weights is used to select forwarding paths for large flows,which can avoid conflicts of elephant flows caused by untimely updating of network resource information.In addition,a rerouting mechanism is added to re-schedule the elephant flow on the link with the highest link utilization when the network load is unbalanced.The simulation result show that this mechanism can effectively improve network throughput and reduce transmission delay.Secondly,in view of the problem that the Dynamic Load Balancing(DLB)algorithm selects the next hop with link utilization,and therefore fails to select the global optimal path,this thesis proposes a dynamic load Dynamic Load Balancing and Energy Efficient Mechanism(DLBEE),which based on the DLB algorithm and combined with the data center energy saving mechanism.DLBEE introduces a method of probabilistic selection of the next hop to reduce the probability that the link is selected with too high or too low utilization,and then sleep the switch in the idle state.In addition,a rerouting mechanism is proposed to reroute the flow on the switch with a low utilization rate to accelerate the speed of sleep and realize energy efficient.The simulation results show that DLBEE not only improves the energy-saving efficiency,but also performs good in load balancing.