Re: .html? File Extensions Perhaps Not So Harmful

On Sun, 14 Oct 2001, Mark Nottingham wrote:

> I don't think this merits 'screw up'. Proxy caching is an
> optimisation mechanism; most don't support Vary in that they don't
> cache anything that is content-negotiated. They will still service
> the request, and indeed handle the Vary header correctly.

Yes, in theory. In practice, when a bandwidth-starving network admin
notices that some bandwidth is wasted because of the Vary headers,
they will make sure that Vary header is ignored or otherwise violated
because it screws up their preferences on how bandwidth is used. As I
am sure you know, people do much worse things to save a few bytes of
HTTP transfers...

When your uplink is 56Kbps, and you have hundreds of users, a caching
proxy is not just an optimization mechanism. People will violate
protocols if protocols affect their bottom line and can be violated.
 
> If a proxy doesn't support Vary in the sense that it will go ahead
> and cache the entity without considering Vary in the cache key, it's
> broken. That's their fault, not the content provider's.

I did not assign faults. A "we will use this cool feature despite it's
actual support or effect" attitude is protocol-legal, but not the kind
of behavior I would recommend.

And, if you want to assign faults based on HTTP/1.1 requirements, keep
in mind that most proxies out there either speak HTTP/1.0 or are not
HTTP/1.1 compliant.

Alex.

> On Thu, Oct 11, 2001 at 09:30:23PM -0600, Alex Rousskov wrote:
> > On Thu, 11 Oct 2001, Ian B. Jacobs wrote:
> > 
> > > Now that SVG is a Recommendation, we'll probably start publishing
> > > several versions of an image, so that you get the best format your
> > > browser supports.
> > 
> > .. and probably screw up most caching proxies out there because they
> > still cannot handle Vary headers [correctly].
> > 
> > Alex.
> > 
> 
> -- 
> Mark Nottingham
> http://www.mnot.net/
>  
>

Received on Monday, 15 October 2001 12:01:57 UTC