Comment on "A Model for CDN Peering"

[Barron Housel <housel@cisco.com>]

The following comments are based on the version dated 9/21.

General comments
    I think this paper should be folded into the architecture document.  Terms in isolation scream for more structure and relationships between the entities etc.

Section 2.

• Do we want to limit CDNs only to "Web-based network elements" at this point?
• Does the statement of "delivery of any digital content" also include dynamic (i.e., program generated) content?  This might be possible if the CDN had the ability to run programs against replicated read-only data.  If this isn't intended, you should use a more restrictive word than "any."
• Last para: replace ";" with ":" after distribution use commas to separate the problem phrases.

Section 2.1

• Should you define "forward proxy" and "reverse proxy"?
• 2nd para: eliminate "and its associated infrastructure" as it adds no information.
• 3rd para, last line: should it be "objects are based" instead of "objects based" ?
• Is another limitation of existing techniques is that most don't discriminate by content provider?

Section 2.2

• Another attribute of "good" is that the content be delivered with the appropriate level of integrity and consistency.
• 2nd para, line 3: eliminate "and" before "second".
• 4th para, line 3: change "speed" to "reduce"

Section 2.3

• 2nd para, line 3: withe -> with

Section 3.

• I would change the title to "CDN Model Terms".
• CONTENT and CONTENT DATA UNIT: Wouldn't life be simpler if you defined CONTENT as a typed finite sequence of octets and eliminate CONTENT DATA UNIT? Then explain that the delivery method may vary by type to capture your continuous media requirement.  Would a "CONTENT" ever consist of more than one DATA UNIT?  The mixing of being able to parse the content in some cases while stating that the structure of the content is opaque is a little confusing.  Do you imply that in some cases the CDN can perform transcoding or media format conversion?  If so I don't think you want to go there.
• I would change "DIRECTION" to "DIRECTING".  I've made this comment before with no success and promise not to mention it again!

Section 5.

• Change "Peering Model" to "Peering Model Terms"
• BILLING CDN and DISTRIBUTING CDN:  Doesn't this confuse things? i.e., There are other billing relationships such as between CDNs.  If publisher has a negotiated relationship with CDN X but not CDN Y but CDN Y has a negotiated relationship with CDN X then CDN Y can bill CDN X for its services and CDN X can pass the commensurate charges to the publisher.  Are these terms really important for describing the architecture?

Section 6.1

•  On a DISTRIBUTION ANNOUNCEMENT pertained to the service levels between the CDN surrogates and a collection of client addresses:  Is content delivery generally sensitive to client addresses (except for determining proximity)?  I would have thought that the client has some delivery agreement (e.g., silver, gold, platinum) with the content provider and the level of delivery would be a function of the agreement between the CDN and the content provider and the type of object to be delivered.  Same comment for ACCOUNTING ANNOUNCEMENT.  I may be missing something here.

Regards,
    Barron Housel