Machine Learning Integrated Software Defined Networking Architecture for Congestion Control

Data center networks rely upon the excellent throughput of the network to ensure data transfer happens at incredible speeds. Many factors can determine the network’s throughput, such as the design, routing mechanism, and congestion control model. The introduction of the cloud made it more challenging as high traffic can happen and disrupt the data transfer for the cloud applications; this is a challenge in hosting cloud services and applications. The introduction of Software Defined Networking (SDN) made it possible to program the networks and thus provide scope for enhancing the congestion control mechanisms. Separating the control plane from the data plane helps in independence and extracting and processing the data from different perspectives. This paper proposes a Machine Learning approach implementable in an SDN context that helps reduce congestion control by learning data transfer patterns. We propose a modified version of SDN in which an additional layer is implemented above the control plane called ML plane. ML plane takes care of the trend learning in the network that guides the control plane for better decisions. In this paper, we exhibit the architecture of ML plane for the purpose of congestion control in busy networks.