- From: Greg Wilkins <gregw@intalio.com>
- Date: Sat, 2 Aug 2014 09:57:35 +1000
- To: Jeff Pinner <jpinner@twitter.com>
- Cc: Jason Greene <jason.greene@redhat.com>, HTTP Working Group <ietf-http-wg@w3.org>
Received on Friday, 1 August 2014 23:58:05 UTC
Jeff, in that example, there are only 7 headers, 10 if you count the request line ones. Of those 6 will probably stay pretty constant over multiple requests, 1 might do and only 3 will change on most requests. Of those, only 1 has a custom header name, which looks constant over time. Thus for requests like these, you are looking at around 20 requests on a connection before the 2 custom header name with constant values are pushed into the 2 byte index range. The one that is constantly changing will constantly be in the low index range as it will freshly be added every request. So I'm still not seeing how 2 byte name index's will be that common with requests like that. Even if they are, they will not be until after 20 or so requests, by which time the slow start windows are opening up and compression is less important. But data is king - so perhaps you can get a sequence of real requests from some real users and run the numbers. I'd be surprised if the 2 byte name indexes were significant, but your mileage may vary. cheers -- 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 Friday, 1 August 2014 23:58:05 UTC