CDNP naming

["Phil Rzewski" <philr@inktomi.com>]

Greetings. I am taking the opportunity to summarize some of the past
naming issues and hope to get some discussion going on the topic. I
apologize in advance for the length of this.

Some of you may remember a post I made several weeks ago regarding my
concern about the interchangeable use of the terms "CDN
Peering" and "Content Peering", which seemed to imply they
were equivalent:

-----
It therefore comes down to the use of the term "CDN". There's
once again a "crowd" that has been using the term in a
particular way, in this case the crowd being industry. Outside of the
IETF/draft communities, there's reasonably well-defined differences
between what's been called CDNs, Hosters, and Access Providers. Now,
doing a meticulous reading of [draft-day-cdnp-model-02], I get the
impression that perhaps there's an attempt to categorize Access Provider
cache deployments as being a degenerate form of CDN, but I think this
might be a mistake because it would create a disconnect between industry
terminology (which was established first) and standards 
terminology.
-----

I went on to describe a particular case of Content Peering that involved
entities like Access Providers. Dan Li described this as a type of
"vertical peering", which sounds like a fair label. Fred
Douglis also seemed to agree that the term "CDN Peering" may be
too specific to describe all scenarios, but he didn't feel too good about
the term "Content Peering" either. Ted Hardie had previously
checked in with a general feeling of uneasiness about the term
"peering", since it often implies a settlement-free model when
applied in a BGP context. I was a little lacking in good ideas for terms,
so I was hesitant to press the issue until I had something more
significant to offer.

Then I posted the OACP and related drafts
(http://www.ietf.org/internet-drafts/draft-rzewski-oacp-00.txt),
which, for this discussion, serve one main purpose: They try to better
articulate an example of Content Peering that involves networks that
would not be called CDNs using today's industry terminology.

I had a conversation with Mark Day yesterday, and he agreed that the
naming is important, but we both felt that to risk letting the topic
dominate the BOF would be bad. Hence my attempt to hash out the ideas
now.

I'll try to pick apart each term:


"Content":
Fred Douglis correctly made the point that using the term
"Content" alone (as in, "Content Peering") is
inaccurate because you're not peering the content itself, you're peering
entities that deliver content (or, as the CDNP drafts correctly state,
these could also be entities that do request routing for content, or
accounting, or a subset containing only one or two of these).


"CDN":
As I had stated previously, I am nervous by the fact that industry has
already defined the term CDN quite strongly. I'll name names to make this
perfectly clear: Networks operated by Akamai and Digital Island have most
surely come to be known as CDNs... I hope nobody from Akamai, Digital
Island, industry analysts, or anyone else would dispute that. Meanwhile,
networks operated by AOL and Prodigy most certainly are not currently
known as CDNs... and I have a sense that people from those companies,
industry analysts, and others would like to avoid creating any confusion
around that. I'm particularly sensitive to this one because, as the day
comes when a formal name might  be assigned to an IETF WG, I fear
that a WG with "CDN" in the title would imply either: 1) The
group only would work on topics relevant to networks that are today known
as CDNs, or, 2) The group would work on topics relevant to multiple types
of peered content networks, and would therefore be on a crusade to
redefine the term "CDN". I would personally like to avoid
seeing either of these.

Mark Day mentioned to me that this group originally titled their work as
being in "Content Peering", but switched to "CDN
Peering" for reasons like the ones that Fred Douglis stated.
However, as he pointed out jokingly, nobody's ever 100% sure if the
"D" in CDN stands for "Delivery" or
"Distribution". Indeed, to take this a step further, the CDNP
drafts talk about peering of entities that, say, only do Accounting. So
they're not doing any "D" of either kind... but I do think
those accounting entities certainly belong in the drafts.


"Peering":

Ted Hardie's point is a good one, that many people are used to thinking
of "peering" as a settlement-free concept, due to how it has
traditionally been implemented in BGP. In response to this, I would like
to say that, in my day running Internet backbones, we did always have the
concept of "peering with settlement" in the back of our minds.
We never set it up in our own network (we had enough trouble doing
usage-based billing for our own customers, why spend time on doing it for
peers?!) but we always figured it might be out there somewhere (after
all, people talked about it enough). As a result, we could just think of
ourselves as implementing to a model where "peering with
settlement" is the norm rather than the exception. For that matter,
I don't see why you COULDN'T implement settlement-free peering under the
CDNP model.

As a result, I have no problem with the word "peering". That
being said, if enough people dislike it, I have no problem throwing it
out for something else. Compared to the term "CDN", I don't
think the industry is as attached to the terms "Content
Peering" nor "CDN Peering". The people I've seen using
these terms the most are analysts trying to predict the next-big-thing
while having no clue what they're really talking about. :)


Ok, so now some proposals. Choose your favorites or add your own:

- If we don't like "Peering", we could replace it with
"Internetworking". (Once again, my vote is that
"Peering" is ok, but I'm not religious on this one.)

- Rather than just say "Content", say "Content
Layer". That is, "Content Layer Peering" or "Content
Layer Internetworking". Granted, there's no such thing as the
"Content Layer" in OSI, but I think we may agree that we're
seeing the emergence of "overlay" technologies that all assume
they begin life at the upper layers of the OSI stack. Therefore, if we
consider the Content Layer as an abstract place where all the protocols
reside that are potentially used for distribution, delivery, accounting,
& injection of content, then Content Layer Peering would be the
cross-network functionality of systems that operate in that layer.

- I see a few alternatives for "CDN". My preference would be to
come up with a more general name for the purpose of the higher-level
drafts. Once again, dropping the "D" makes room for things like
networks that only do Accounting. One way to do this would be to just use
the term "Content Networks": A network that somehow "does
stuff" with content. That way, things like ISPs' farms of forward
proxies are also "Content Networks", and if they peer with a
CDN to get content signals (e.g. invalidations), they're participating
under the model. If you want to be able to cover accounting,
distribution, delivery, & more all in one term, where's the shame in
being general? OR, if you want to talk specifically about Content
Networks that wanna do stuff across administrative boundaries, then you
could augment the ones from the last bullet to make terms like
"Content Layer Internetworking Entities", or "Content
Layer Peering Entities". These seem too wordy to me though. I'd vote
for "Content Networks".

Ok, enough outta me.... for now. :)

--

Phil Rzewski - Senior Architect - Inktomi Corporation                 

650-653-2487 (office) - 650-303-3790 (cell) - 650-653-1848 (fax)