RE: Distribution CPG Protocol - Some Thoughts

[Stephen Thomas <stephen.thomas@transnexus.com>]

At 09:06 PM 2001-01-04 -0500, Oliver Spatscheck wrote:
>I agree with you that we won't need per transaction real time feedback for the
>purpose of direction (we might need it to fulfill legal requirements for on
>line gambling ....), however, we need fairly fast feedback of the sort CDN A
>is covering network B in region C with available capacity D for service E.

Maybe this is just a red herring, but the notion of "regions" can come up 
several times on this thread, with the implication that we're referring to 
geographic regions. Am I the only one that's bothered by this? I'm bothered 
for a couple of reasons. First, is the notion of terrestrial geography even 
relevant here. No doubt some clue-challenged folks at a large content 
provider would be pleased to hear "the CDN we just signed up for has 
surrogates in France." At little digging, however, might reveal that the 
major non-PTT ISP in France does not have very good connectivity with the 
PTT ISP, and, in fact, most of the traffic between the two ISPs actually 
routes through the United States. (This is just a made-up example, but it 
is representative of reality; the last time I saw the statistics, over 40% 
of trans-Atlantic Internet traffic was actually being routed to the US and 
immediately turned back around and returned to Europe.) If the content 
provider really wanted to improve the experience of French users (75% of 
whom, say, use the PTT ISP), the provider would be better off using 
surrogates in the US rather than surrogates in the non-PTT's network. The 
second thing that bothers me is how we represent regions in any kind of 
protocol. I suppose we could use ISO two-letter country codes, but that's 
just binary. (You're either in fr or you're not.) There's certainly no 
algorithmic way to determine proximity. E.164 country codes (for telephone 
numbers) might do a little better, since the first digit generally 
represents a region, but there are a lot of worms in that can.

Would we be better off referring to autonomous systems rather than regions 
or countries? It seems more relevant to the problem at hand, and there is 
an unambiguous identifier. Of course, algorithmic determination of 
proximity is still a problem.


____________________________________________________________________
Stephen Thomas                                       +1 770 671 1888
TransNexus, Chief Technical Officer    stephen.thomas@transnexus.com