Benefits for LWN subscribers: Subscribing to LWN helps keep it publishing. Subscribers get immediate access to all site content and extra features. Sign up today!

QUIC:

• A UDP-based transport protocol forming the foundation of HTTP/3, developed at Google in 2012 and an IETF standard in 2021. The QUIC Working Group published several follow-up standards and is working on four more extensions.
• Serves as a replacement for TCP and TLS with benefits like reducing round trips, hiding connection details from routers, and supporting multiple independent streams.
• Load balancing: Connection migration allows computers to change IP addresses while maintaining a connection. Load balancers face challenges as QUIC packets contain a connection identifier. The QUIC Working Group is working on a new standard to securely encode load balancing information in the connection ID. Martin Duke said the draft is mostly ready but on ice.
• Multipath: TCP still has an advantage in multipath support. The in-progress QUIC multipath extension would adapt the connection-migration mechanism. There are complications with network address translation rebinding and packet numbering/encryption. The draft has gone through many revisions and is still under development.
• Acknowledgment frequency: QUIC needs to handle data integrity and congestion control. Currently, acknowledgments are sent for every second packet. The acknowledgment frequency extension allows participants to request changes. It is theoretically ready but the authors have not moved forward.
• Partial delivery: QUIC allows separate streams to fail independently. The QUIC Working Group is addressing this with a draft extension to define a RESET_STREAM_AT message. The draft is nearly ready for publication.
• Conclusion: These potential improvements promise to make QUIC more useful and performant, but progress has been slow. Christian Huitema expressed frustration with the working group's progress. It's uncertain how long it will take the IETF to finalize the new standards, while QUIC adoption is growing. One of the key concerns is avoiding protocol ossification.