RE: Distribution CPG Protocol - Some Thoughts

[Oliver Spatscheck <spatsch@research.att.com>]

Stephen Thomas writes:
 > At 08:49 AM 2001-01-05 -0500, Oliver Spatscheck wrote:
 > 
 > 
 > >Since I introduced the word region I would like to clarify that
 > >I never intended to imply geographic region. Internally we use
 > >the term region as a collection of prefixes.... . I think AS
 > >is to coarse grain. If I tell you it is in 701 what do you
 > >really know. So I guess neither term is intuitive and clear.
 > 
 > 
 > I'm not real happy with AS either, mainly because there's not algorithmic 
 > relationship between different values. (ASN 701 and 702 may be no "closer," 
 > in a network sense, than ASN 701 and 3383811.)
 > 
 > By prefixes do you mean IP address prefixes? If so, is that any better than 
 > AS numbers? Maybe within a subnet. (There's a pretty good chance that 
 > 172.16.1.1 is real close to 172.16.1.2.) But would you say the same thing 
 > outside of a subnet? (Which is closer to 172.16.1.1 network-wise, 
 > 172.17.1.1 or 10.1.2.1?) In fact, I'd think that ASNs might actually be a 
 > little better since they don't have the ambiguity of variable netmasks. 
 > 172.16.1.1 and 172.16.2.1 might be on the same subnet (if the netmask is 16 
 > bits), but they might also be 50 hops away on the other side of the world 
 > (if the netmask is 24 bits).
 > 
 > 
 > 

Internally we define a region as:

REGION : <NAME> {
       <IP>,<PREFIXLEN>;
       <IP>,<PREFIXLEN>;
       ....
}!

where all prefixes included in a region are close to each other using a
predefined metric and threshold.  We do not imply that prefixes which are next
to each other numerically have any particular relationship.  However, we
started to define relationships between regions. So having structure in
a region name becomes helpful.

ASs are also a poor choice since they are huge. AS 701 covers the entire US
and has many POPs. I would like to be able to select the CDN with a cache closest
to the POP my customer is connected to.


Oliver