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

On 16 July 2014 23:30, Tatsuhiro Tsujikawa <> wrote:

> ​From the table, h2-12 shows best efficiency to me.

Well to be precise h2-12-noRefSet-noCopy-static1 showed the best
efficiency, but now h2-12-noRefSet-noCopy-static1-date5 does.

However, I think the data set is way too small to be claiming any of these
(other than date5) gives the best result.   Looking at the results from the
individual stories in the test data, they are each +/- 5%. I have simply
combined the results from all stories and there is no rigour in that as
some are probably more frequent that others.

So reporting the figures to two decimal places was not a valid thing for me
to do and the following table probably more accurately represents the
inaccuracies of this test:

HPACK/HEADER TABLE01024409616384h2-1264%47%42%42%h2-12-noRefSet64%47%43%42%

Ie, none of these variations (except date5) give any significant savings on
this data set.   Hence I think the simplest is the best:

But I would be very interested to see a concrete proposal for


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

Received on Wednesday, 16 July 2014 13:58:50 UTC