RE: CDNP naming

[Gary Tomlinson <garyt@entera.com>]

On Wed, 6 Dec 2000 at 6:67 PM Eric Dean wrote:

>Just to be clear in what my concerns are here.  If I'm a PUBLISHER and my
>CDN provider goes off and peers with all sorts of other CDNs then I may
>have a few problems:
> 
>If all those CDN's do not have equivalent capabilities then I may be
>forced to break up my content into little descriptive elements using DNS
>as the only method. For example, I may have to creatively encode my
>objects using a various hostnames like
>   
>small_objects.cdn.net
>large_objects.cdn.net
>gzip_objects.cdn.net
>stream_objects.cdn.net
>.
>.
>.
>
>How else will a CDN, receiving only a DNS query, know whether to route the
>request to a peered CDN with sufficient capabilities or not?
>

It has been the intent of many members of the original design team to
address these issues.  The architecture draft explicitly calls out the need
for a uniform URI name space encoding of meta data.  Several of us imagined
policy information pertaining to object types, service types and capability
expectations would be encoded in such a manner, that DNS or other
request-routing mechanisms could make good coarse request-routing decisions;
at least to a suitable CDN.  

Additionally the architecture allows for the coupling of the distribution
peering system with the request-routing peering system for the purposes of
communication of important request-routing hints, such as service
capabilities, load, proximity, etc.

I've been assuming we'd get into genrating requirements for these once the
WG gets going.

Gary