Continuing the Discussion...

[Stephen Thomas <stephen.thomas@transnexus.com>]

Taking an opportunity to summarize some of my thoughts.

What do we want to accomplish? My own (undoubtedly skewed) view is that a 
majority of participants, though not necessarily a consensus, was 
comfortable with the notion that we want a standard interface to content 
distribution networks. Whether the interface is used by content publishers 
or peer CDNs is mostly irrelevant, so long as the interface is not 
significantly biased in to favor one role or the other. In the interest of 
simplicity and focus, I would suggest that we do not consider the interface 
between a client and a CDN. (Several sub-reasons: first, one could argue 
that no such interface should exist, rather that the existence of a CDN 
ought to be transparent to the end user; second, the examples cited in 
discussion-client wants to indicate a particular quality of service-is 
probably better handled by the client telling the publisher its preferences 
and having the publisher reflect those preferences back to the CDN.)

Although I did not sense a consensus (yet?) on this, in my own opinion a 
minimum interface to a CDN requires at least two exchanges. First, the 
publisher must describe the content to be delivered. Second, the CDN must 
return accounting information to the publisher. Either or both of these 
could be accomplished off-line, but I think eliminating either from the 
scope of work reduces the problem so significantly that it is not worth 
solving. Also, I'm being deliberately vague by merely saying "describe the 
content", although I do think that naturally includes attributes of the 
content such as TTL. Again in the interests of simplicity and focus, I 
would suggest limiting the description to attributes of the content. I 
would explicitly exclude any discussions of attributes of the distribution 
network (e.g. don't store this content on a surrogate in the UK because the 
publisher doesn't want to be subject to the R.I.P. law) or of attributes of 
the client (e.g. password, credit card number, ...). Both of these areas 
may be fruitful follow-on work, though.

Where do we want to accomplish this? The fact that the IETF is being 
considered as the venue for this work suggests that there is at least a 
suspicion that the interface will be implemented as a communication 
protocol. Although the IETF does more than define protocols, that is 
certainly its strength. If the prevailing view is that the CDN interface is 
more appropriately documented as a data structure (e.g. an XML DTD), then 
perhaps another body (W3C?) would be more appropriate as a venue for the 
work. (Note that I am personally not yet convinced that defining a protocol 
is the right approach here; it might be useful to have some further 
discussion about the merits of, say, a DTD.)


____________________________________________________________________
Stephen Thomas                                       +1 404 872 4887
TransNexus, Chief Technical Officer    stephen.thomas@transnexus.com