Leveraging SDN Paradigm for Dynamic Load Balancing through Minimum Latency Link for Data Center Networking

Abstract

In modern digital era, conventional Data Center Networks (DCN’s) are not good enough to accommodate huge number of users and their bandwidth requirements that are constantly changing. Evolution of big data, cloud computing, Internet of Things (IOT), block-chain, quantum computing demands a preferable networking infrastructure to handle network expansion and network usages fruitfully. In DCN, bundling of control and data plane in same networking device limits its functionality for dynamic computation and storage access. In general, traditional networks operators or administrators directly employ the load balancers that are vendor specific and nonprogrammable. The load balancing developed in traditional network infrastructure is not precise as it is based on local information of the network. Due to existing static routing mechanisms in traditional networks, most of the network resources are still underutilized. This dissipate of network assets is becoming common in today’s traditional typical networks. Software Defined Networking (SDN) emerges as a new platform that promises to control, change and manage the inherent services of networking nodes by extracting statistics from lower layers of the topology facilitating network engineers and administrators. Load balancing is most renowned application of software defined networking that offers fairly load share between network nodes, optimizing best path along with bandwidth and reducing latency. SDN offers a global view of whole network at one place, a centralized controller while helps in making satisfactory and upright decisions. SDN promotes flexibility, programmability and refined manageability. The load balancers designed through SDN platform are programmable, optimized, scalable and non-proprietary. It means everyone can implement their own load balancer depending upon their own requirements. These are the software load balancers different from the hardware load balancers adopted in traditional networks. In our project, floodlight SDN based controller was chosen for implementation of dynamic load balancing. The Dijkstra’s algorithm is exercised in our application running on controller. A data center network, fat tree topology of open flow switches is deployed to depict real life traffic complexity in data center networks. Different Application Programming Interfaces (API’s) are used to gather information regarding all the devices connected in the topology including their network and hardware addresses, link and switch latency’s, bandwidth with least cost path for data packets. To create virtual topology of nodes, MININET emulation platform is utilized. Different load balancing verification mechanisms justify that our load balancing technique is doing a splendid piece of work.