RE: Executive summary of: RE: hard questions: request routing

["Scanlon, Paul " <Paul.Scanlon@qwest.com>]

Again the problem of a single metric, the number of milliseconds to the
client's DNS has no bearing in an in-line model ("- Request-routing
protocols MUST support DNS-based and in-line request-routing systems.")...I
would tend to lean towards the restrictive case as a model that seems to
offer a plausible level of success.  However, within in this model would
there be an application of policy to facilitate the ability to direct a
request?  I don't see the requirement for it in the draft, is there going to
be a way to apply policy based preferences for request direction?  Perhaps
that mechanism will come after this discussion has concluded...

- Paul

-----Original Message-----
From: Eric Dean [mailto:eric@crystalballinc.com]
Sent: Wednesday, April 04, 2001 8:15 PM
To: Dmitri Krioukov
Cc: Mark Day; cdn@ops.ietf.org; Oliver Spatscheck
Subject: RE: Executive summary of: RE: hard questions: request routing



A metric need not represent anything specific.  If everyone wants it to be
the milliseconds to a client's DNS proxy then so be it but it's not a
requirement.

On Wed, 4 Apr 2001, Dmitri Krioukov wrote:

> I see the point about consensus but I don't see why
> something without deep semantics (like number of CDN
> hops (ala RIP metric)) has not been considered as
> "obtainium" :) Oliver's p.1 is just a specific
> case with infinity equal to 2 then.
> --
> dima.
> 
> > -----Original Message-----
> > From: owner-cdn@ops.ietf.org [mailto:owner-cdn@ops.ietf.org]On Behalf Of
> > Mark Day
> > Sent: Wednesday, April 04, 2001 10:42 AM
> > To: Dmitri Krioukov; Oliver Spatscheck
> > Cc: cdn@ops.ietf.org
> > Subject: RE: Executive summary of: RE: hard questions: request routing
> >
> >
> > > I missed the point when the metric was dismissed, though :(
> >
> > I haven't yet seen any convergence on what the single common metric
would
> > be. I tend toward agreement with Oliver's summary that we seem to have
> > implicitly dismissed that by our inability to determine the metric.
> >
> > The single common metric seems to be an example of what my aerospace
> > colleagues refer to as "unobtainium" -- a wonderful substance
> > with all sorts
> > of great properties, which unfortunately doesn't exist.
> >
> > While it can be useful to do unobtainium-based exercises, at some
> > point you
> > have to start translating those designs into realizable ones.  Unless
> > someone believes that we actually approached rough consensus on a single
> > underlying metric and I missed it (which is quite possible!), I'm
inclined
> > to say that we have spent enough time on architectures that depend on
this
> > particular form of unobtainium.
> >
> > --Mark
> >
> 
> 

Eric Dean
President, Crystal Ball Inc.
W 703-322-8000
F 703-322-8010 
M 703-597-6921