Fixing HTTP/2 and preparing for HTTP/3 over QUIC

Deploying HTTP/2 correctly can be challenging in practice, and HTTP/3 will make things even more difficult as the underlying QUIC protocol runs over user datagram protocol (UDP). Robin Marx explores practical proxying, caching, load balancing, and routing issues and how to overcome them.

HTTP/2 has been standardized for four years now, and uptake is high. However, as recently highlighted by Andy Davies and Patrick Meenan, many multi-tier and cloud-based HTTP/2 deployments are badly configured, which leads to severe performance degradation as the HTTP/2 prioritization system fails to function correctly. Robin Marx takes a concrete look at the issues and explains the technical reasons, detailing how to test your own setup’s behavior and how to mitigate any problems you might encounter with actionable solutions. And after making sure your HTTP/2 setup is performing well, you’ll look to the future in the form of HTTP/3 and QUIC. As QUIC runs on top of UDP (as opposed to transmission control protocol (TCP)) and is fully end-to-end encrypted at the transport layer, deployments can encounter a slew of issues in terms of firewall, cache, proxy, load balancing, and routing setups. As not all networks will allow UDP, it also becomes necessary to run concurrent HTTP/2 and HTTP/3 stacks so users have a fallback option. Robin walks you through the QUIC and HTTP/3 protocol internals and explores the provisions that help to deal with these issues (e.g., the connection ID). You’ll discover concrete approaches that large companies like Facebook, Google, and Cloudflare are taking in their setups, and you’ll see the next generation of QUIC and HTTP/3 tools that can help you test your cutting-edge setup. And Robin touches on aspects such as 0-RTT connection setup and resumption. You’ll leave able to tweak and validate the performance of your HTTP/2 stack and start practically preparing and testing for the HTTP/3 madness.