Towards a Learning-Based Framework for Self-Driving Design of Networking Protocols

• Published in: IEEE access Vol. 9; pp. 34829 - 34844
• Main Authors: Pasandi, Hannaneh Barahouei, Nadeem, Tamer
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 2021; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Approximation algorithms; Communication protocols; deep learning; IEEE 802.11 Standard; Machine learning; machine-generated algorithm; Media Access Protocol; Optimization; Parameters; Protocol; protocol design; Protocols; Reinforcement learning; Throughput; Tuning
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2021.3061729

Networking protocols are designed through long-standing and hard-working human efforts. Machine Learning (ML)-based solutions for communication protocol design have been developed to avoid manual effort to adjust individual protocol parameters. While other proposed ML-based methods focus mainly on tuning individual protocol parameters (e.g. contention window adjustment), our main contribution is to propose a new Deep Reinforcement Learning (DRL) framework to systematically design and evaluate networking protocols. We decouple the protocol into a set of parametric modules, each representing the main protocol functionality that is used as a DRL input to better understand and systematically analyze the optimization of generated protocols. As a case study, we introduce and evaluate DeepMAC a framework in which the MAC protocol is decoupled into a set of blocks across popular 802.11 WLANs (e.g. 802.11 a/b/g/n/ac). We are interested to see which blocks are selected by DeepMAC across different networking scenarios and whether DeepMAC is capable of adapting to network dynamics.