Carrier Sensing Transmission Technique Study for Performance Improvement in Optical Data Center Networks

Multiple applications based on Artificial Intelligence (AI) and Machine Learning (ML) emerge with an accelerating rate, requiring that various cloud Data Centers (DCs) around the world can process a constantly increasing number of datasets. As a result, the DC networks (DCNs) are requested to serve enormous number of tasks in an effective way. In this framework, the optical switching technology seems to be the most promising one for the DC servers interconnection grace to its flexibility and effectiveness, as compared with the electrical one. In this paper, we investigate the access control requirements in optical DCNs targeting their performance optimization and aiming to effectively serve traffic high bandwidth demands like those of AI applications traffic. Especially, we propose an efficient Carrier Sense Multiple Access with Collision Detection (CSMA/CD) protocol, called Congestion Sense Medium Access Control (CS-MAC) protocol, for optical DCN environments exploring the bandwidth utilization level, as compared to the conventional CSMA scheme. Our proposal is based on the sensing of the congestion conditions in the network in order for the CS-MAC protocol to force the servers to properly adjust a lower number of transmissions, aiming to guarantee sufficient bandwidth to overcome the network congestion. Limited analytical study combined with simulation results show that the suggested CS-MAC protocol provides almost 100% bandwidth utilization, much higher than the conventional CSMA scheme, being a promising solution for the service of multiple AI and ML tasks traffic in optical DCNs.