Re: Call for Consensus: Remove "reference set" from HPACK (to address #552)

I ran the hpack test data through the various options being discussed for
several different header table sizes:

HPACK/HEADER TABLE01024409616384h2-1264.22%47.06%42.18%41.82%h2-12-noRefSet

There is actually very little variation in results +/- 1% which is likely
to be below the variation due to different data sets.       So on that
basis we should go for the simplest mechanism, which is either
h2-12-noRefSet-noCopy  or h2-12-noRefSet-noCopy-static1

But if we trust that data set, then the best compression was achieved with
h2-12-noRefSet-noCopy-static1 and a large header table.    Also
h2-12-noRefSet-noCopy-static1 was better than h2-12-noRefSet-noCopy for all
but the 1024 table size.

So on this data we should go with h2-12-noRefSet-noCopy-static1

h2-12 == the HPACK used by draft 12 of HTTP
noRefSet == no Reference Set
noCopy == Do not copy static fields to the header table
static1 == Static indexes are 1-61, dynamic are 62-...


On 16 July 2014 11:01, James M Snell <> wrote:

> On Tue, Jul 15, 2014 at 1:21 PM, Martin Thomson
> <> wrote:
> > On 15 July 2014 12:14, Jason Greene <> wrote:
> >> Whatever happened with this idea? Searching my archives I couldn’t find
> it getting rejected. I also don’t see an issue filed for it.
> >
> > Search for "typed codecs" and variations on the same.
> >
> > Also:  (note
> > the version number)
> >
> Note also that the version number is reflective only of my own
> editing/publishing style and not of any particular amount of
> consideration from the WG
> - James

Greg Wilkins <> HTTP, SPDY, Websocket server and client that scales  advice and support for jetty and cometd.

Received on Wednesday, 16 July 2014 02:14:17 UTC