Adapting reinforcement learning for multimedia transmission on SDN

Abstract

Multimedia transmissions require a high quantity of resources to ensure their quality. In the last years, some technologies that provide a better resource management have appeared. Software‐defined networks (SDNs) are presented as a solution to improve this management. Furthermore, by combining an SDN with artificial intelligence (AI) techniques, networks are able to provide a higher performance using the same resources. In this paper, a redefinition of reinforcement learning is proposed. This model is focused on multimedia transmission in an SDN environment. Moreover, the architecture needed and the algorithm of the reinforcement learning are described. Using the Openflow protocol, several sample actions are defined in the system. Results show that using the system users perceive an increase in the image quality three times better. Moreover, the loss rate is reduced more than half the value of losses recorded when the algorithm is not applied. Regarding bandwidth, the maximum throughput increases from 987.16 kbps to 24.73 Mbps while the average bandwidth improves from 412.42 kbps to 7.83 Mbps.

Machine Learning/AI for IoT, M2M, and Computer Communication

“…The following four papers deal with the application of ML in SDN and network virtualization. Benayas et al presented autonomous SDN fault management platform that uses Bayesian diagnosis in the seventh paper, “A semantic data lake framework for autonomous fault management in SDN environments.” Article 8, “Adapting reinforcement learning for multimedia transmission on SDN,” by Rego et al, enhances decisions of OpenFlow and guarantees QoS by applying ML on multimedia transmissions of SDN. The ninth paper, by Zangiabady et al, “Self‐adaptive online virtual network migration in network virtualization environments,” leverages reinforcement learning to minimize the cost associated with migrating virtual network resource from one to another.…”