W3C home > Mailing lists > Public > ietf-http-wg@w3.org > July to September 2012

Re: Would a header/schema compiler help?

From: David Morris <dwm@xpasc.com>
Date: Thu, 2 Aug 2012 14:30:47 -0700 (PDT)
To: "ietf-http-wg@w3.org Group" <ietf-http-wg@w3.org>
Message-ID: <alpine.LRH.2.01.1208021419430.22206@egate.xpasc.com>


On Thu, 2 Aug 2012, Phillip Hallam-Baker wrote:

> Question,
> 
> In the case that a server and client do the HTTP connection keep-alive
> thing, does the client send the cookies on the first request only or
> for each request?
> 
> If we have MUX, that may be where the performance gain comes.

The client must send the cookies, UA, etc. on every request as their is no
state preserved between requests.

The accept headers can get quite bloated as well.

There is no reason for most headers to be repeated if the protocol
provided for reuse of previous headers. I lean toward explicit reuse,
such as a 'mark' of some kind in the re-using request or response
to indicate a reference to a previous header.

I don't advocate simple compression (LCZ et al), but I'm surprised
by the finding that only the UA and Cookie headers made a difference.
I would expect stream compression (e.g., keeping the dictionary open)
to reduce most headers to almost nothing after a very few R/R pairs.
Also, using an initial dictionary trained with header names,
common UA terms, accept tokens, etc. should make a big difference.

Dave Morris
Received on Thursday, 2 August 2012 21:31:16 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Thursday, 2 August 2012 21:31:22 GMT