- From: Greg Wilkins <gregw@intalio.com>
- Date: Fri, 25 Jul 2014 09:20:28 +1000
- To: HTTP Working Group <ietf-http-wg@w3.org>
- Message-ID: <CAH_y2NEgmD18rCDVfJKGB7mCraswUqrCLBLvfCfF5TWGyRBGeA@mail.gmail.com>
In the test data there are 38037 fields in 3267 headers of which 161 have
duplicate header names:
58 Duplicate: cache-control
43 Duplicate: content-type
19 Duplicate: pragma
19 Duplicate: expires
9 Duplicate: set-cookie
7 Duplicate: x-content-type-options
2 Duplicate: last-modified
2 Duplicate: content-length
2 Duplicate: accept-ranges
However, I think this says more about the quality of the data set than it
says about the likelyhood of duplicate headers - duplicate cache-control
and content-type fields are almost certainly not good.
For my part, I'm not that interested in the byte saving of avoiding
encoding a duplicate name. It is more work to discover if the name is
duplicate than to invest a few bits to just encode it and move on.
cheers
On 25 July 2014 04:30, Roberto Peon <grmocg@gmail.com> wrote:
> I'd suggest keeping it unless we have data suggesting it isn't necessary.
> It isn't exactly the most complex of things.
> -=R
>
>
> On Thu, Jul 24, 2014 at 11:26 AM, Martin Thomson <martin.thomson@gmail.com
> > wrote:
>
>> On 24 July 2014 11:03, Roberto Peon <grmocg@gmail.com> wrote:
>> > I believe that it would make sense to potentially (at the compressor's
>> > option) have same-named header fields referenced by a single op-code.
>> > Analysis may prove that this is a needless concern though.
>>
>> What do you suggest that we do? Based on the information we have on
>> the two options, the reduced complexity suggests that '\0'-concat
>> isn't worth it. Unless you would like to find some more information
>> to tilt the scales in the other direction. I'd suggest that would
>> need to happen quickly, just as I suggested the same to Herve when he
>> suggested we use ranges for header table references (though that would
>> require tuning in addition to tests).
>>
>
>
--
Greg Wilkins <gregw@intalio.com>
http://eclipse.org/jetty HTTP, SPDY, Websocket server and client that scales
http://www.webtide.com advice and support for jetty and cometd.
Received on Thursday, 24 July 2014 23:20:57 UTC