Re: Reference set in HPACK

[Roberto Peon <[e-mail]>]

Kazu--

Even a few bytes per request adds up as the number of requests go up.
Lets think about a page today that has 100 elements, which is something we
see today.
If the headers were regularized, with the use of a reference set, one could
imagine a reduction of 20 bytes per header.
With 100 elements, this is approximately 2k of data, or two packets worth.

This makes the difference between being able to use server push as a
workable optimization technique or not being able to do so since it might
require more packets/space than inlining, and thus present a disadvantage
as compared to inlining for the first, cold, pageload.

The trend is to increase the number of individual resources on a page.


Note that a single byte is used only when the number of table entries is
fairly small. This is likely to be the case for client<->server
communication, and less likely to be true for proxies.
-=R


On Tue, Jul 1, 2014 at 10:52 PM, Kazu Yamamoto <[e-mail]> wrote:

> Hi Roberto,
>
> > You're basing conclusions on today's data, instead of looking forward as
> to
> > what might happen when the set of headers sent adapts to the compression
> > method, making it significantly more likely for items in the reference
> set
> > to be emitted.
>
> If you believe that reference set is suitable for headers in the
> future, please give me such examples so that I can convince.
>
> In my understating, reference set cannot contribute for ANY headers.
> Typically an index is represented by one byte. Even if you can omit
> sending some indices, we can save just some bytes.
>
> --Kazu
>