Comment From: mgravell
But RESP isn't even remotely an HTTP thing. Can you clarify?
Comment From: mitchsw
RFC 9000 (QUIC) is finally published, and basically makes no reference to HTTP. Using QUIC's concept of streams could be helpful for RESP to implement true multiplexing without the challenges of opening 50+ TCP connections and/or pipelining in the application.
Comment From: madolson
Multiplexed incoming streams are a pretty cool concept for Redis, especially for clients that are multi-threaded and have multiple connections.
I assume that the requestor is really just looking for QUIC, and not HTTP3 (Redis only supports TCP). This would be cool if someone wanted to investigate this more deeply and understand what types of changes we need to make it order to make it happen. I would guess this isn't a super high priority for us right now.
Comment From: yossigo
@madolson The first place we'd probably want to see stream multiplexing is replication links, where this could solve the issue of replica buffers altogether.
Comment From: madolson
@yossigo Yes, that would also be useful, but that is more of an internal infrastructure improvement and not really a direct end user feature.
Comment From: madhub
Redis protocol implementation on top QUIC should help improving performance by * Reducing overhead during connection setup ( including TLS) * Improve overall latency and throughput.
Other application level protocols already experimenting in this area Here is how Microsoft implemented SMB protocol over QUICK
Comment From: masx200
any update？
Comment From: madolson
Redis became a source available, so a lot of the folks that were working on it are no longer here. Feel free to re-open to https://github.com/valkey-io/valkey.