SMASH: Supervised ML-Based Adaptive Selection of HTTP Protocols

This paper presents a novel system and method for optimizing the selection of HTTP protocols, specifically QUIC and TCP, using machine learning techniques to enhance the performance of User Equipment (UE). Our observation reveals that the performance of these protocols can vary based on factors such as region, network configurations, and the user equipment's settings. The primary objective of this study is to evaluate the performance of HTTP/2 (TCP) and HTTP/3 (QUIC) in real-world conditions and propose methods for selecting the most suitable protocol dynamically. Our focus is on comparing the speed of QUIC connections with TCP and providing insights into the adaptive selection of HTTP protocols. We introduce the Supervised ML-based Adaptive Selection of HTTP Protocols (SMASH) approach, which leverages machine learning techniques to dynamically choose the optimal HTTP protocol for applications. To implement SMASH, we introduce an eBPF -based layer to manage the flow of QUIC packets, providing granular control over the communication process. Major blockchains today work on TCP-based protocols. SMASH will enable blockchain applications to improve the connectivity parameters. Our experiments demonstrate that SMASH aims to improve connectivity between applications and servers by achieving a significant performance boost of approximately 20 - 30%. By dynamically adapting to network conditions, user equipment configurations, and regional variations, SMASH ensures faster and more efficient communication, contributing to an enhanced user experience in live-air conditions.