Re: HTTP router point-of-view concerns

[Nicolas Mailhot <[e-mail]>]

Willy Tarreau <w <at> 1wt.eu> writes:

> 
> On Sat, Jul 13, 2013 at 06:48:31PM +0000, Poul-Henning Kamp wrote:
> > It could actually be interesting if somebody would go through their
> > HTTP/1.1 traffic and estimate how much bandwidth would be saved.
> 
> On most static servers (ie most of the round trips for loading a page),
> it will actually *inflate* the traffic since we normally don't send
> cookies there, but with the proposal, I suppose the session ID will be
> sent anyway (or we need at least one bit to indicate its presence).

How about this:

1. servers mark the parts of a web site that need a routing label (could be
some form of url extension, for example @ without login or @@ in presence of
login)

2. they propose a new routing label to clients that connect to such a part
without one

3. the client then answers either with this label or the one it used
previously (if it remembers it)

4. the client can refuse to reuse a label already allocated to another site
(not same origin or different certificate, perhaps, policy is client-side)

5. there are strong protections against routing label abuse :
– lifetime limited to two hours since last exchange (enough to handle most
interruptions and lunch breaks)
– lifetime limited to 12 hours since set up (a big work day)
— size limited so you can not stuff other things in there

-- 
Nicolas Mailhot