Deepsdn: Connecting The Dots Towards Self-Driving Networks

The cloud data centers are going through an unprecedented growth from past few years. In an era of real-time video streaming, on-demand gaming, door-step e-commerce services, and highly inter-connected social networks, cost-effective service models, adaptive resources provisioning and upfront applications availability contribute significantly towards such a stellar growth. However, there are many challenges that must be addressed in a systematic manner to meet the requirements of increasingly demanding current and upcoming applications of the cloud computing paradigm. Optimum resources management, instant response time, interoperability among a diverse set of emerging technologies and innovative applications are a few of these challenges. On the other hand, the recent trend in softwarization of networks, particularly enabled by network function virtualization (NFV) and software-defined networking (SDN) principles, provides immense opportunities to better utilize the network resources by programmable abstractions with an efficient control and management techniques. Furthermore, machine learning based solutions are gaining prominence in resource optimization problems and autonomous systems. Therefore, in this paper, we strive to connect the dots by state-of-the-art methodologies in networking and machine learning domains and utilize these developments to grapple with the challenges of the cloud-based systems. We propose DeepSDN, an SDN-based solution that harnesses existing machine learning techniques to move a step closer towards self-driving networks. The comparative results obtained from an experimental testbed corroborates effectiveness of our approach and suggest a way forward towards autonomous network management