Re: Query re request routing

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

At 09:00 AM 3/5/01 +0000, Steve Rudkin wrote:
>In the case of broadband streamed media on-demand, surrogates need to be as
>close to the edge of the network as possible and content will mostly likely
>be pushed to the surrogates in advance.
>In this situation, I would therefore expect that
>- with respect to any given request, all candidate surrogates will be within
>the same CDN, (referred to as the destination CDN)
>- the origin could well be operated by a different CDN though, (referred to
>as the source CDN)
>
>Whilst there is a need to interconnect distribution systems of the source
>and destination CDNs (and any intermediaries) and to do the same for
>accounting systems,  is it necessary for the RRS of the destination CDN to
>interact with other RRSs? If so, why? If not, should the statement that
>there is a single authoritative RRS for each *content unit* be
>modified/qualified in some way. Would it be better to say there is a single
>authoritative RRS for each *request* ?

I agree with this distinction. It's one of the types of things I am trying 
to capture in Scenarios by enumerating something called an "Access Content 
Network" (ACN).

As I see it, there are always going to be Access Content Networks that will 
try to "own" the user's session. This has traditionally been access ISPs 
(such as those that use transparent interception to redirect people to 
caching proxies--yes, naughty, but very real) and in the future will 
probably be the broadband-type ISPs to which you refer. These "first 
content hop" boxes (called "avatars", to use an OPES term), owned & 
controlled by that ACN, are there for the enhancement of the user's entire 
session. A fine example is the avatars that are in the cable head-ends of 
cable modem providers today. With the CDI model, these avatars become 
likely candidates to receive content pushes as you say, and they then 
become "surrogates" in that context (since they're doing the bidding of 
some origin server by receiving distributions and sending back accounting 
data). But the request-routing is completely handled within the domain of 
that ACN. For this model to work, you can assume that the terms of the 
"negotiated relationship" essentially say that the ACN will be doing all 
the necessary capacity planning on those avatars in order to maintain an 
adequate level of service for their user base. By signing the contract, the 
publishing side basically assumes the ACN will maintain that level of 
quality. The publisher is free to use measurement systems to verify this, 
but they're not like they have some actual control method (in the form of 
authoritative request routing) to ever stop delivery happening from those 
ACN's avatars.

In summary, the ACN is doing some form of "edge caching" for their user 
base today, and they're going to keep doing it. Since that "first content 
hop" is always going to be owned by that ACN, that will always be a 
potential bottleneck to the user experience (though it won't be if the ACN 
does their job). So the publisher might as well accept that fact and 
leverage the fact that it's there (make the most out of it: leverage it for 
distribution, milk it for accounting data).

There's still plenty of room for the CDI model we've worked with up until 
today (that the Authoritative Request-Routing system, near the publisher, 
has ultimate control). That's because there's going to be plenty of 
Internet "edge" that won't be covered by avatars, but will be covered by 
CDNs. There might be two CDNs with surrogates in that neighborhood, it's 
just a case of choosing the "best" at that moment.

--
Phil Rzewski - Senior Architect - Inktomi Corporation
650-653-2487 (office) - 650-303-3790 (cell) - 650-653-1848 (fax)