[Dundi] advertising for peers

[Joe Abley]

Hi all,

I've read through the archives on this list, and some of the 
preconceptions about peering suitability seem a little out of whack 
(but then, perhaps I just need to be illuminated, so bring on the 
headlights).

1. It makes no sense to me to impose regional conventions on telephone 
numbers on a system which is designed to connect regions, not provide 
connectivity within them. So, every reference to regionalisms such as 
area codes, LATA, NPA, NXX should be removed from peoples' minds, and 
they should instead concentrate on advertising collections of E.164 
numbers ("routes", right? I'm a packet guy, normally).

If I am an enterprise to which calls in the 100-number block 
+165042313xx are routed, then the information I need to publish is 
"+165042313", or maybe "+165042313xx".


2. If I'm looking through a list of potential peers, I don't think it 
makes sense to have them listed geographically 
(Country/State-or-Province-or-Whatever/...). What I think I want to do 
is analyse my telephony costs, and if I notice I'm terminating a lot of 
calls on +441234, I want to hunt for peers that can service those 
numbers.


3. The decision as to whether entity A should peer with entity B seems 
like it is all about the relative, respective usefulness of the routes 
available to those two entities, and has very little to do with 
designations of "tier-1", "ISP", or notions of available bandwidth.

Available bandwidth, in particular, says nothing about the capacity 
along a path through the Internet between two dundi implementations. 
Juggling routes according to network performance (whether due to 
long-term congestion or short-term maintenance or network failure) is 
surely part of the everyday business of connecting telephones together; 
if calls towards a peer network are consistently bad, you re-route them 
elsewhere. And if they're consistently good, there's no reason to 
assume they won't be bad tomorrow.

So, I'm not sure I understand why "bandwidth" has any relevance at all. 
If the idea is to make the core of a peer-to-peer graph nice and dense 
so that the average distance for lookups is managable, there are surely 
better metrics that can be used.


Having said all that, here are some details, and we'd like to hear from 
other people who would like to experiment with this stuff. (Advanced 
warning: we likely have some Asterisk upgrades to complete before we'll 
be ready to send any packets.)

Internet Systems Consortium, http://www.isc.org/

BANDWIDTH: 2+ Gbit/s external connectivity, 7 transit providers, ~100 
peers
PUBKEY: on all the usual key servers
COUNTRY: US
STATEPROV: CA
CONTACT: jabley at isc.org
CONTACTNAME: Joe Abley
FAX: +1 650 423 1355

ROUTES: +1 650 423 13

We have staff in three continents, equipment and network deployed in 
five, and we regularly host public conference bridges with participants 
from many countries, both for ISC-sponsored activities and for external 
groups (ICANN, IETF, etc).


Joe