At 19:56 10/10/00 -0400, Keith Moore wrote:
> >
> > CDNP
> > ====
> > * Distribution of data from origin servers outside of the administrative
> > control of the data owner
>
>assuming I understand what you mean by the above, I don't think IETF
>should be endorsing such practices at all. I see two pieces:
>
>1. how the content provider makes data available to consumers, and
>how it makes the locations of those data known to consumers
>
>2. how consumers find and access such data
>
>the first is under control of the content-provider; the second
>is under control over the consumer.
>
>I see no legitimate role for a third party that isn't acting under
>explicit instructions from one or the other of the first two
>parties. IMHO, *this* is what IETF should standardize.
Mark was right in his comments. This was meant to mean distribution to
servers where the data owner/content provider is not the owner of the
systems to which the end-user connects in order to retrieve material (e.g.
surrogates within a CDN).
> > (I'd specifically exclude the scenario where the distribution occurs
> within
> > a single administrative domain... do we really need to standardize that?)
>
>yes, because there is a need to have standardized interfaces between
>content providers and content-distribution networks acting on
>behalf of the content providers. there is also a need for a standardized
>and efficient interface between content users and content
>distribution networks - the stuff we are using now is quite crude.
Looks like I was misunderstood, since I meant that part to be where a
content provider wanted to do distribution across systems that they owned
rather than CDNs. Is that something that we should be looking at?
> > * Routing request to closest site (service) with the requested data
> > - should probably consider caching proxies in the model
> > - (obviously has some aspects of competitive advantage as was discussed
> > from the WREC KP draft in Pittsburgh)
> > - definition of the "closest" site (see WREC KP draft on known failings
> > of ICP in heterogeneous network)
>
>not sure how to define it better but "closest" (or any superlative) may
>be over-ambitious
Yes, that was the sort of point I was trying to raise. The KP draft gives
an example where "close" (network RTT) may be worse when there's more
bandwidth available between 2 points.
We need a succinct way to say "there, yes, that thing, that's in the
currently optimal place to serve it". I seem to recall some heated
"discussion" over the use of the word "optimal" in Mark's presentation at
WCW5 earlier this year.
This archive was generated by hypermail 2b29 : Thu Nov 18 2004 - 11:21:29 MST