Reducing HTTP payload size [was: RE: HTTP idea

First, I would note that in general, HTTP headers are such a small
percentage of the HTTP bytes transfered that the potential ROI is likely
discourage either design effort and subsequent adoption. But if reduction
of HTTP headers were desirable, a general mechanism which provided for
collaborative caching of selected headers in the context of a persistent
connection would be more useful than focus on accept headers. Something
like:
   Please remember this list of headers and values under this name as
   I expect to use the same values multiple times on this connection.

Then a request (or response) could reference previous headers as a set.
Lot of details would need to be worked out ranging from how to recognize
when a server (client) supports header caching, possible over-ride of
specific cached values, etc.

Another potential alternative would be to revisit some variation of a
binary protocol such as was being investigated as http-ng.

Finally, I would observe that without serious attention to more efficient
representation of content, there is really nothing we can do at the
protocol level efficient representation wise that will not be totally
eclipsed by the raw size of the textual representation of the page as the
move continues to richer interaction models.

OTOH ... some of the evolving uses of HTTP requests to obtain small bits
of data to be filled in by client side logic, validate single fields,
etc., might run much faster with terser HTTP requests ... but then even
much of that value would be achieved by having browsers aware of
programatic interactions and optimizing the headers actually needed. But
in that case, using XML to encode those requests might still negate any
advantage of smaller protocols.

Bottom line ... a lot of thought should be given to ROI and adoption
issues before spending much effort defining more efficient protocols.

Dave Morris


On Thu, 11 Jan 2007, Larry Masinter wrote:

>
> I don't know if we might want to say more about the practical
> limitations of using accept: headers in HTTP, but see:
>
> http://www.alvestrand.no/pipermail/ietf-types/2006-April/001707.html
>
> and the follow-up discussion, including
>
> http://www.alvestrand.no/pipermail/ietf-types/2006-April/001722.html
>
>
> Larry
>
>
>

Received on Thursday, 11 January 2007 17:37:50 UTC