Analyzing the impact of QUIC protocol on a WebRTC call using machine learning

Quick UDP Internet Connections (QUIC) protocol has been gaining popularity due to its low-latency high-throughput transport layer capabilities. To date, QUIC has been implemented by a limited number of platforms (e.g., Google Meet, YouTube, etc.) and it has been standardized with RFC9000. Limited academic work explores the impact of QUIC on the application Quality of Experience. This is a challenging problem for two reasons; there are many versions of QUIC and limited ways of extracting the application performance statistics. In this work, we performed an experimentation-based analysis of QUIC on WebRTC. We first constructed a data set from WebRTC traces across cellular and home Wi-Fi access networks, some using QUIC. Next, we trained machine learning algorithms on this data set to predict WebRTC freeze count from application statistics. We comparatively analyzed the impact of using QUIC protocol on the WebRTC performance and couldn't find that QUIC has a positive effect on predicting freeze count over HTTP/2. To the best of our knowledge, this is the first work that experimentally investigates the impact of QUIC on the performance of WebRTC-based applications.