Re: Distribution CPG Protocol

[Alex French <afrench@vianetworks.com>]

At 08:56 PM 12/01/2001, Stephen Thomas wrote:
>Sounds to me like an argument for just shipping around a bunch of generic 
>values. If that's the case we can just co-opt, say, SCSP [RFC 2334] and be 
>done with it. Or maybe the two parties could just hack together a couple 
>of CGI/Perl scripts and run them on a couple of Linux/Apache boxes. Any 
>competent

I agree with Stephen; in addition, completely generic metrics will make it 
almost impossible to set up transitive relationships across CDNs. To go 
back to the BGP metaphor, BGP does have some "generic" attributes, but also 
quite a few specific ones.

To elaborate on why generic metrics cause problems with transitive 
relationships, imagine a content provider (CP) that contracts to a CDN (A) 
for distribution. A in turn has contracted with another CDN (B) to provide 
distribution in particular networks (or countries).

So CP and A agree a set of metrics offline; say for example footprint, 
aggregate bandwidth and minimum reserved space on each surrogate. But A may 
have a different set of metrics in its peering with B, for example instead 
of aggregate bandwidth, B may only agree to negotiate minimum bandwidth at 
a set of IP peering points. In general, it is impossible to translate 
between these two set of metrics, which will make it impossible for A to 
give CP an SLA.

In practice, if we are to have properly transitive CDN internetworking, all 
CDNs will have to end up using the same set of metrics. So let's just try 
and figure out what they're likely to be, and leave some wiggle room for 
"custom" metrics in case we're wrong.


--
Alex French
Consultant, Technical Services                     E: afrench@vianetworks.com
VIA NET.WORKS, Inc.                                T: +353 (86) 818 8118
12100 Sunset Hills Rd, Reston VA 20190      F: +353 (865) 818 8118