Research On Routing Technique Of Data Center Networks Based On SDN

With the rapid growth of network traffic,network operators continue to deploy more and more servers and storage devices,resulting in an exponential growth in the amount of nodes and links in data center network.The network complexity is inevitably increased with the increases of network equipment due to the close relationship between network management and network physical topology in the conventional network architecture.Then,the network management becomes more and more difficult and it restricts the improvement of network quality of service seriously.Software defined network(SDN)is a new type of network architecture,which can simplify the network devices by decoding between the control plane and the forwarding plane,and realize the programming ability for the switch directly.It also provides simple and efficient configuration and management for the upper layer by abstracting the network and shielding the underlying complexity.Simultaneously,the SDN controller can obtain lots of information including the topology information,the status information and the application requirements for the global network in real time.Therefore,it is a convenient way to select the optimal path for the network flow.Nowadays,it is a hot research topic in studying the data center network routing strategies and algorithms based on SDN,especially in carrying out in-depth research on effectively solving the problem of network load balancing and network energy consumption.This thesis starts from the research on the traffic characteristics of the data center network based on the SDN architecture and its implementation protocol—OpenFlow.And then it goes to the design of two specific routing mechanisms depending on problems of network load balancing and network energy saving based on the centralized control programming of SDN architecture.The details are as follows:(1)We designed a dynamic load balancing routing mechanism based on the flow classification of SDN in distinguishable flow characteristics size of network traffic in data centers.Firstly,the developed mechanism collects the network topology,switch load and other information through OpenFlow protocol in order to route flow dynamically.Then,the mechanism detects the potential elephant stream only for reducing the controller overhead.Finally,it reroutes the elephant stream to select the maximum available bandwidth path to avoid collisions.In this thesis,we simulated two flow models under Fat-Tree topology.Results show that the proposed routing scheme can improve the average throughput and load balancing performance of the network and the load distribution on the core switch is more uniform compared to the traditional ECMP and Hedera dynamic programming routing schemes.(2)We proposed the optimization for the energy saving in link utilization constraints based on the traffic characteristics of day and night mode in data centers.Also,we designed the bandwidth-aware energy-saving routing mechanism based on SDN.Under the conditions in maintaining the network connectivity and network maximum link utilization constraints,the scheme makes the traffic concentrate on links as few as possible by sleeping the switches and the links,leading to achieving energy-saving targets while balancing the network load and avoiding network congestion.In this thesis,we simulated the random flow model on Fat-Tree and BCube topologies.Results show that our developed algorithm succeeds in balancing between different energy consumption values and different congestion thresholds,reducing the network energy consumption and optimizing the load distribution.Compared to the greedy algorithm of ElasticTree scheme,the proposed energy saving scheme is better at low loads.