Seems like this only helps large (heavy) websites with consistent content. The real world examples are all large, like YouTube, Amazon, etc...
Small JSON responses that compress to <1k would fit in a single packet, so I don't see the advantage of going from "65 bytes with normal Zstandard compression, vs 28 bytes when using the past response as a dictionary - 57% smaller."
by dmitrygr
1 subcomments
So this is just LZ with a pre-populated window? Any backreferenceing compression can be used this way - just prepopulate the backreference history on both client and server up front and off you go. Why is this new?