Challenges and Limitations in Automating the Design of MAC Protocols Using Machine-Learning

To cope with the emergence of new technologies, various device characteristics and application requirements, complex and custom design of high performance networking protocols is much needed. Networking protocols, practically, are designed through long-time and hard-work human efforts. However, these designed protocols, typically, have limited flexibility that results in non-optimal performance under several network scenarios and conditions. Therefore, replacing this inefficient human based designing process by a novel paradigm that enables rapid design of efficient, flexible and high performance protocols that intelligently adapt to different device characteristics, application requirements, user objectives, and network conditions is highly desired. In this paper, we motivate the importance of a shift from human-driven protocol design process to a machine-based design. We propose a novel, self-managing and self-adaptive framework for automating MAC protocol design. As an example of such a framework, We design, implement, and evaluate AlphaMAC framework that learns to automate the design of efficient simple MAC protocols. We decouple MAC into a set of building blocks, and we are interested to see what blocks are selected by AlphaMAC in different scenarios, and whether the designed protocol is efficient. Our results show that AlphaMAC is able to select the efficient set of building blocks from ALOHA protocol building block set such that the designed protocol outperforms conventional ALOHA. We also discuss some of the challenges and limitations of realizing such a framework. We believe that the impact of the automated design of networking protocols on the network research and industrial community, and on developing networking services and applications would be significant.