2.1.17 Content Distribution Internetworking (cdi) BOF

Current Meeting Report

Content Distribution Internetworking (CDI) BOF Notes

Taken by Paul Knight; paknight@nortelnetworks.com

The meeting was called to order at 1530, Monday August 6, 2001.

Phil Rzewski opened the meeting with agenda bashing. There were no changes to the published agenda, which was simply a draft-by-draft review.

Phil gave an introduction for first-timers. He described the focus, emphasizing that it is connectivity between Content Delivery Networks (CDNs), not the internal operation of individual CDNs. He described the "black box" principle for CDI, which states that CDNs participating in a CDI arrangement should not be required to reveal any internal topology details.

>From the viewpoint of a content user, the CDI must operate in a "reasonably transparent" mode, so that the user does not need to know the process by which the user is connected to the eventual content source (surrogate).

Phil discussed the group activity since the last IETF meeting.

Mark Day presented the "model" draft report. The draft has been revised to coordinate with RFCs 2616 (HTTP) and 3040 (Internet Web Replication and Caching Taxonomy). It has also incorporated language from various previous drafts. The main task for other drafts in CDI is to adopt the terminology of "model." There should be only little tweaks remaining before moving to last call with "model." There were no questions.

Phil presented the "scenarios" draft report. There has not been much response on the -03 draft. We need some consensus on new content network (CN) types. These will be rolled into "scenarios" unless there is some objection. The scenarios will be prioritized in a -04 draft. Phil requested input from service providers on "scenarios."

Question/statement from floor: Are there plans for addressing CDI requirements for mobile users? This may be applicable for distribution of content to automobiles.

Answer: This is interesting. Please discuss this on the mailing list.

Gary Tomlinson presented a report on the "architecture" draft. It will be aligned with the terminology of "model." There were no questions.

Abbie Barbir presented the "known request routing mechanisms" draft report. The terminology has been aligned with the "model" draft. He requested input from the CDI community on the mailing list.

Question (Ingrid Melve): DNS-based CDI approach only seems to support specific source [?] addresses. What about anycast addresses?

Answer: That may need to be addressed. Let's discuss that on the mailing list so we can get wider input.

Question (Ingrid): What about security considerations?

Answer: Since this draft reviews existing mechanisms, the security section will only mention any well-known security shortcomings of these existing mechanisms. It is not the place to address CDI security issues. We will add a security section to the draft.

Brad Cain reported on the "request Routing requirements" draft. The major remaining work appears to be synchronizing with the "model" terminology. However, there is some discussion of combining this with the "distribution requirements" draft because of interdependencies between the two. If that is done, then there will be major rewriting of the draft.

Question: (John [last name ?] ) Interception proxies are precluded from joining a CDI arrangement without authorization from the content source? Does this include non-interception proxies?

Answer: Anything not authorized by the content source should not be able to join the distribution/delivery system.

Question: (John) How about end-user authorization?

Answer: let's discuss in the mailing list. One example is a corporate proxy server.

Question (John): We need to be clear - who is controlling the proxy - this may raise some privacy issues.

Answer: Yes, we need to add more on this issue. However, this will probably not explicitly discuss forward proxies.

Question: Does the draft specify that there will be a single exit point [?] from one CDN to another CDN?

Answer: Yes, this is the CIG.

Question: (Dan Rothman) If there is a need to authorize a content delivery (from either the source or the end-user), what is the mechanism to replicate or forward the authorization?

Answer: This is systemic. We need to clarify the role of proxies, which are not interception proxies. Is this where to address authorization?

Statement (Gary Tomlinson): We will add proxies to the "architecture" draft. Also security, including authorization.

Question: What is the relationship of the CIG with middleboxes, as discussed in MIDCOM?

Answer: (Brad) Interworking can be extended for multiple services. Some people consider CDNs to be middleboxes.

Question: The architecture appears to have no interface for a Content network (CN) being the content provider (origin). Consider the Digital Rights Management (DRM) issues. This can bring things down. Need to provide a CN-to-origin interface. Answer: In the drafts, we try to treat the origin as just another inter-CN interface.

Mark Day presented a report on the "Accounting requirements" draft. This may be renamed to reflect that it covers AAA, not just Accounting.

Question: Was the scope of CDI AAA too big to start with? Maybe it should just cover billing, not cover everything beginning with "A".

Answer (Mark): (review of history of draft) There was a request to align with AAA. It's not clear what is the minimum needed to be successful to support CDI Internetworking.

Question: Could you let it go for now, and define it later, assuming initial connections will have strong business relationships?

Answer: This is not on the same time frame as other documents. Lisa Amini presented a report on the "Distribution requirements" draft. She highlighted several interdependencies with other CDI drafts and with WEBI. She described several issues, including support for streaming media, partial objects, security, and DRM implications. She asked for feedback and comments on the mailing list.

Question: This needs reconciliation with RUP. RUP is intended for the content space. Answer: (Lisa) Yes, the draft should explicitly point to RUP, but in its current form, RUP does not fully encompass the CDI requirements. For example, the RUP document has recently been modified to say it is only focused on invalidation. Regardless, I will work on having the document reflect what is being covered by RUP. The key is to separate the requirements for the distribution and payload components, in the hopes that the distribution component of the RUP document could indeed cover the CDI requirements. The payload would probably remain separate, and specifically cover the CDN requirements for initiating/updating/withdrawing content (distribution/delivery) services (not invalidation).

Mark Day discussed the need for an editor to work on the Accounting/AAA draft. He asked for volunteers. Dan Rothman stated that he is interested, and will be able to decide whether he will be able to be the draft editor within two weeks. Pat Calhoun also expressed some interest in working on the draft, since he is involved with DIAMETER.

Question: (Fred Douglis) The CDN mailing list has had very little traffic recently, no discussion. What's going on? Is the work being done in "smoke-filled chat rooms"? Answer: (Mark Day) I believe it's genuinely quiet and not a result of private conversations not making their way to the list.