Re: h2 header field names

On 5 September 2014 00:52, Poul-Henning Kamp <> wrote:
> Also I don't recall ever seeing the actual process used to derive
> the huffman table documented, and suspect that it should be documented
> and reviewed before we finalize the huffman table.

It should be in the minutes of the NY interim.  Chrome deployed an
experiment that performed HPACK compression on all requests and
responses seen by their browser over a significant data set.  The
resulting counts for each octet were collected.

That was before removing the reference set removal so results will be
dominated slightly more by header fields that change frequently and
there will be instances of '\0' (used at that time to ensure ordering
of values was maintained) that are no longer valid.  I can speculate
that a new experiment along the same lines would shift the bias
slightly in favour of base64 characters, but only a similar experiment
will decide that.

(I'll also note that the results fortuitously produce sequences that
don't have unique mappings from length to character, which reduces
information leakage from the resulting compressed length. The research
Google sponsored indicates that this is minimal.)

Received on Friday, 5 September 2014 16:10:51 UTC