The QUIC Fix for Optimal Video Streaming

TL;DR

This paper proposes extending QUIC to support unreliable streams, demonstrating that such an approach can significantly improve video streaming performance over traditional TCP and QUIC implementations.

Contribution

It introduces a simple method to add unreliable streams to QUIC and shows how this enhances video streaming efficiency and performance.

Findings

Unreliable streams in QUIC improve video streaming performance.

The prototype outperforms TCP and standard QUIC in evaluations.

Supporting unreliable delivery benefits network operators and content providers.

Abstract

Within a few years of its introduction, QUIC has gained traction: a significant chunk of traffic is now delivered over QUIC. The networking community is actively engaged in debating the fairness, performance, and applicability of QUIC for various use cases, but these debates are centered around a narrow, common theme: how does the new reliable transport built on top of UDP fare in different scenarios? Support for unreliable delivery in QUIC remains largely unexplored. The option for delivering content unreliably, as in a best-effort model, deserves the QUIC designers' and community's attention. We propose extending QUIC to support unreliable streams and present a simple approach for implementation. We discuss a simple use case of video streaming---an application that dominates the overall Internet traffic---that can leverage the unreliable streams and potentially bring immense benefits to network operators and content providers. To this end, we present a prototype implementation that, by using both the reliable and unreliable streams in QUIC, outperforms both TCP and QUIC in our evaluations.