Research On The Openflow-based Load-balancing Routing In Distributed Networks

Nowadays, with the constant development of the computer networkingtechnology, especially for cloud computing, the network operatingenvironment is also changing faster and faster. Thanks to the boom intechnology, the cost of computers and storage devices slumps. However, itbrings us a new issue, i.e. to improve the IT infrastructure and it’s nowbecoming a top priority. No doubt this also applies to network realm. Wefind it more and more difficult to carry on innovative experiments based onthe existed network equipment and protocols. In recent years, the emergenceof OpenFlow technology opens a fresh chapter for the innovative research innetworks. It gradually becomes the keyword to lead the next generation ofnetwork revolution. In data center networks, how to balance the workloadsis a key issue with the fast growth of network applications. OpenFlowprotocol, which is a competitive candidate for solving the problem, provides each user the programmatic control for specific flows, so as to determinetheir paths through a network. However, existing solutions based onOpenFlow only try to find a static routing path during initialization stepwhile the static routing path often suffers from poor performance since thenetwork configuration may change during the data transmission, or sharpchange in link states may take place.To solve this problem, we will first propose a novel single-hopload-balance routing algorithm (LABERIO) to timely and globally balancethe load in the distributed network with centralized control. It works well inintegrating and utilizing the information of the entire network, monitoringthe usage of global resources by the central controller and specific algorithm,thus dynamically selecting new routes for flows. In this case, when theload-balance detector exceeds the one-hop LABERIO trigger threshold, thecentral controller will be notified and help schedule the biggest flow on thebusiest hop. Besides, for each flow, to decide which path to take initially, weapply the max-min remainder capacity strategy (MMRCS), i.e. to pick outthe path of maximum available bandwidth from all minimum remaindercapacity on each possible path, and set it as the initial path selectioncriterion. However, this single-hop routing method has some notableshortcomings. For example, when the network topology is far from full-populated or if the available substitute paths are too long compared tothe original one, we can foresee a surge in the path switching cost.To deal with the shortages mentioned above, we proposed anotherdynamic routing approach, called multi-hop LABERIO algorithm. Whensome link are over-loaded, we find out the entire biggest flow, and substituteit with a new one. During the data transmission, we scan the usage of eachlink at a certain frequency. Of every interval τ, the overloaded hops(measured by the reminder bandwidth on them) will be picked out, if exist,and be put into a set. Then we find out the flow which covers the most hopsin Σ and set it as our object flow (the biggest flow) for path switching. Thisapproach achieves good performance when applying to the fat-tree topologynetwork.To sum up, this paper proposed two approaches of dynamic loadbalancing according to two specific network topologies respectively. Both ofthem fully utilized the specialty of OpenFlow switch and protocol. Besides,they outperform other typical methods, including LOBUS and Round Robin,by reducing the end-to-end transmission time to a large extent. Finally, weput forward experiment results to show the privilege of our algorithms andthe comparison with other existing benchmarks.