Dynamic load balancing in SDN-based data center networks

Abstract

With the migration of computational powers and applications to the cloud, Data Center Networks (DCNs) have become the backbone of the underlying infrastructure. Operation of the data centers relies on huge computational resources and bandwidth, that often undergo high operational costs, frequent link congestions, and imbalanced traffic loads. Software Defined Networking (SDN) based traffic load management improves accessing resources by distributing traffic among multiple paths efficiently and in a timely manner. In data centers, SDN-based traffic management techniques control paths of incoming flows and optimize flows during their transmissions. In this paper, we propose an SDN-based dynamic load management algorithm for optimizing link utilization in DCNs while considering the flow priority. The algorithm finds the shortest paths from each host to others and calculates every link's cost. When congestion occurs in a certain path, it replaces the old path with the alternative best route that has the minimum link cost and lower traffic flow. Performance of the algorithm is evaluated by measuring throughput, delay and packets loss in a fat-tree DCN. Simulation results show improved performance in load balancing over time as the algorithm keeps on running.