Openflow-based global load balancing in fat-tree networks

Cloud services have been explosively popular over the last decade. And data centers play an essential role in providing cloud services. Inside a data center, any server instance has the chance to inject traffic of various applications into the network. Yet how to balance the enormous internal load to make the best of data center network is a highly prioritized problem to be solved. To provide balanced traffic in data centers, this paper proposes an OpenFlow-based GLB load balancing algorithm in data center fat-tree networks. GLB uses a path-related weight to select path. This weight indicates how balanced of a path. We implement GLB algorithm as a module in an openflow controller platform, POX. On the self-defined modified mininet emulation platform, we conduct experiments in a fat-tree topology environment running random traffic to generate performance data. Experiment results demonstrate that our proposed GLB algorithm outperforms DLB algorithm in terms of load balancing.