Re: Straw Poll: Restore Header Table and Static Table Indices

[Willy Tarreau <[e-mail]>]

On Tue, Oct 07, 2014 at 06:23:37AM +1100, Mark Nottingham wrote:
> Thanks, Jeff.
> 
> I see people have already started to respond to this. 
> 
> Everyone else, please do the same ? if you think this needs more discussion,
> please do so, but I think we?re at a point where people can just state their
> preferences.

I'm not strongly in either direction. I think the new design is simpler,
but I wouldn't want to see more failures in field later if we already have
one bad example. I do think that if we could shave one more bit somewhere
to be able to have a cheaper reference to the indexed headers, it would
satisfy everyone, and I'm pretty sure this could be achieved, probably
at the expense of making almost unused encodings more expensive. We need
to keep in mind that the compression has substantially changed since it
was designed and that maybe some combinations are not needed anymore but
their encodings are still assigned, thus could be reused for something
else to gain in efficiency.

I had sent two ideas about this subject that would probably deserve being
revisited here 3 months ago :

      http://article.gmane.org/gmane.ietf.http-wg/23155
      http://article.gmane.org/gmane.ietf.http-wg/23201

In both cases, the point is to be able to encode *at least* some indexed
header fields with low bits, and I think that would help in the case of
people who have many non-common headers. I guess this situation could
become more widespread on WAN links between many enterprises communicating
with web services because there it's common to see a lot of custom headers.

Regards,
Willy