[quicwg/base-drafts] Replace HPACK integer encoding (#1148)

[krasic <[e-mail]>]

Using the same data as an in [earlier thread](https://www.ietf.org/mail-archive/web/quic/current/msg03004.html), I noticed that a decent percent of bytes go to indexed representations, about 10% the example I looked at.    In this particular example, every indexed representation encoded in a single byte.    

If instead of byte-aligning every instruction as HPACK requires, only the entire header block were byte aligned, I could imagine saving two or more bits per indexed instruction (minus some bits of padding per header block),  an average of two bits reduced would be around 2.5% overall savings.    I happen to think this will be a bigger compression win than a lot of the other optimizations in flight*.

Part of my premise here is that for many connections the number of active dynamic table entries is relatively small, on the order of 64 or less.   I haven't sought out the best varint scheme.   A simple starting point for thought, one might construct a static huffman table, where probability is index value, i.e. smallest indices encode with fewest bits.  ( Byte alignment of the whole block might use a few extra bits of padding )

* As in the PR for separate instruction sets in [1141](https://github.com/quicwg/base-drafts/pull/1141).

-- 
You are receiving this because you are subscribed to this thread.
Reply to this email directly or view it on GitHub:
https://github.com/quicwg/base-drafts/issues/1148