Fedora Core brevity vs server upgrades

[Les Mikesell]

On Mon, 2005-05-09 at 20:30, John Summerfied wrote:
> >>
> >>On the other hand, if DNS returned multiple addresses for one name,
> >>the load would be spread among the sites automatically,
> > 
> > Which happens now.

Except that a lot more of it happens than necessary.  I pull about
20 copies of everything myself that previously would have been
cached nicely.  And others update their own systems here too, using
the same proxy.  This is a small office so that's probably not
unusual at all.

> > 
> >>caching proxies
> >>would do the right thing,
> > 
> > I'm sure that would depend on the proxy and your definition of "the right
> > thing".  Caching metadata is a hinderence, not a feature.

The 'right thing' turns out to be the same right thing as for generic
http proxies: get a new copy only if the one on the target site is
newer than the one in the cache.

> Where the machines share an Internet connexion, why would I want them to 
>   refetch the metadata regarding the remote sites? What _I_ want is all 
> machines to use one mirror, one set of metadata, one collection of packages.

More to the point, why refetch if the cached copy is identical?  If you
haven't checked recently, let the proxy do a 'get if newer' and
let the server reply with a '304 Not modified' response like the
http protocol was designed to do...

-- 
  Les Mikesell
   les at futuresource.com