Comments on CDN Peering Architectural Overview

These comments are based on the version dated 10/20.
Barron Housel

General:

Exellent start, very packed with insights and articulates the relationships well for the level of maturity we're at.
As I commented on the model doc, I'd fold in the Model material here to make the document more complete and simplify life for the reader.

Details

Section 1.

Section 2.

Item 3 under the fundamental assumptions: This is not clear. Is this a new proposal for structuring URLs to include meta data? Can you give an example?
On Figure 1 and the example flow: The flow of the request is not clear. Does the URI travel: client->direction peering system->surrogate? or client->direction peering system->(address of surrogate)->client->surrogate? I assumed the latter from looking at the figure; this would be analogous to the way DNS works. Maybe reworking steps 3. and 4. would help. Also, it would be most clear if you labeled the lines (arrows) in the figure and then referenced them in the example.

Section 3.1

Are we constraining the content to be location to be a function of domain name only -- which, in general, can map to multiple locations? Does this relate to the URL meta-data you mentioned earlier?
In the paragraph: "The DIRECTION PEERING systems is hierarchical ..." should PUBLISHER URI really be PUBLISHER domain? (related to previous comment)

Section 3.3

An 8th problem you could mention here: How do we choose among multiple eligible surrogates?
You broach the question of policy distribution in 5. What is the relationship between CDN policy distribution and the other policy work going on in IETF? Is COPS relevant? Maybe a footnote is appropriate to acknowledge a "to do" to look into it.

Section 3.4

1st para: can you give an example of information DIRECTION SYSTEMs may commmunicate?
2nd para: Are we really going to require exchanging potentially a huge number of client addresses? or perhaps subnets? Are there ambiguities introduced via DHCP and NAT in some cases? Isn't the issue whether or not a client can get to content published to the CDN by a presumably paying content provider? Does the content provider provide a list of the clients that can access the content? (doubtful)
Item 2 under Request Direction Requirements: Is URI name space really the DNS name space?

Sections 3.5.4.1 and 3.5.4.2

I have a little trouble with "virtual SURROGATE". Thus far SURROGATES deliver content. The virtual SURROGATE seems to process the request to derive a real SURROGATE. I presume that the real SURROGATE delivers content directly to the client without having to revisit the virtual SURROGATE. How about just calling it a "content-aware redirection network element" or DIRECTION EXTENSION ELEMENT and avoid the confusion. Also, the virtual SURROGATE seems to be outside the DIRECTION system but it is in fact doing request directing.

Section 4.

A general comment on advertising. I expected to see some relationship between the DISTRIBUTION CPG and the DIRECTING CPG. i.e., When content is replicated (distributed) in another CDN, doesn't the location of the content and its URI have to be passed to the DIRECTION CPG in the receiving CDN so the client can locate the surrogate more efficiently? Or does the URI visit the owning CDN's CPG to determine where all the surrogates are?