Re: Distribution CPG Protocol - Some Thoughts

[Martin May <Martin.May@activia.net>]

Oliver Spatscheck wrote:

> Stephen Thomas writes:
>  > At 10:22 AM 2001-01-05 -0500, Oliver Spatscheck wrote:
>  > >Internally we define a region as:
>  > >
>  > >REGION : <NAME> {
>  > >        <IP>,<PREFIXLEN>;
>  > >        <IP>,<PREFIXLEN>;
>  > >        ....
>  > >}!
>  >
>  > If a CDN advertises (or otherwise claims to "cover") a particular IP
>  > address prefix, what exactly does that mean? Does it mean (a) that the
>  > prefix is an atomic subnet, and (b) that there's a surrogate on that subnet?
>  >
>  > Here's an example of where things might get tricky. Suppose I own the IP
>  > addresses 172.16.0.0/16. If I've further subnetted to 172.16.1.0/24,
>  > 172.16.2.0/24, etc., is there any way I can claim 172.16.0.0/16 in the CDN
>  > protocol? Do I have to have a surrogate on all 256 subnets? Just on the
>  > subnets that actually exist at the moment? Or can I just put one surrogate
>  > (say, 172.16.1.1) and argue that no one else is going to get any "closer"
>  > to the other subnets? (Even if 172.16.99.0/24 is connected by a cruddy
>  > radio link that's no better than 9600 baud?)
>  >
>
> I think you are describing one of the important problems we have to address.
>
> My take would be that you can define attributes for regions, with some
> standard attributes like " I own the regions IP address space and have proxies
> " to some attributes which are custom between CDNs. However, it would be nice
> if even those custom attributes could be communicated in our framework.

I think we come to the point where it becomes tricky:
how do we define the "closest" surrogate?
as long as you are an ISP/access provider you know if you are the closest,
but how about the cost?
unfortunately, the "cost" is very important for the choice of
the appropriate surrogate. you may not want the closest, but
the cheapest surrogate.

we should go ahead and define a set of interesting network/server
properties (like number of hops, delay, bandwidth, availability, "cost",...)

Martin


>
>
>  > Is this stuff that we even care about in a protocol? Or should we leave it
>  > open to "good citizenship". After all, the IETF doesn't have a police
>  > force. So even if we said that you had to have a surrogate physically on
>  > the subnet, who will punish the liars.
>  >
>
> I don't think we should care about the enforcment of it. This I would
> say is up to the participating CDNs to figure out. However, we
> should be able to specify those properties and support as much
> as we can the verification of such advertisments somewhere in the
> framework.
>
> Oliver