FlexHTTP: an Intelligent and Scalable HTTP Version Selection System

Mengying Zhou,Zheng Li,Shihan Lin,Xin Wang,Yang Chen

DOI: https://doi.org/10.1145/3517207.3526972

2022-01-01

Abstract:HTTP has been the primary protocol for web data transmission for decades. Since the late 1990s, HTTP/1.1 has been widely used. Recently, both HTTP/2 and HTTP/3 have been proposed to achieve a better experience on web browsing. However, it is still unclear which of them performs better in different scenarios. In this paper, we first leverage the controllable experimental environment of the Emulab testbed to conduct a series of measurements and find that under different network conditions and web page structures, neither HTTP/2 nor HTTP/3 can always perform better. Motivated by this finding, we propose FlexHTTP , an intelligent and scalable HTTP version selection system. FlexHTTP embeds a supervised machine learning-based classifier to select the appropriate HTTP version according to network conditions and web page structures. FlexHTTP adopts a set of distributed agent servers to ensure scalability and keep the classifier up-to-date under network dynamics. We implement a proof-of-concept prototype of FlexHTTP on the Emulab testbed. Experiments show that FlexHTTP achieves a reduction of Speed Index by up to 600ms.