[ppml] [narten at us.ibm.com: PI addressing in IPv6 advances in ARIN]

[vixie at isc.org]

> On today's Internet, settlements are all bilateral.
> The two parties estimate the traffic balance between
> them without actually accounting for every penny of
> cost. Doing it this way, they both save the substantial
> costs of counting pennies. Then, if the both agree
> that they are more or less in balance, they peer 
> with each other at no charge.

and if only one party agrees, you get into cogent/level3
style peering wars.  also, i recall leading a team who
built a global backbone where i could afford to peer with
PSI in every city where they had a presence, which i
offered to do.  they declined, citing not cost-disparity
(since i'd equalized that from my side) but benefit-disparity,
and offered to sell me transit.  apparently this business
model wasn't sustainable for either them or me, given what
happened to AS174 and AS6461 in the year or two following.

so, i think the settlement model described above is somehow
incomplete, and i advise against anyone treating the situation
as being "quite that simple."

> However, if they estimate that there will be a traffic
> imbalance between the two parties, then they agree 
> on a fee to be paid for this imbalance. This is called
> paid peering. 

or if one party thinks they can hold their customers as ransom
and force the other guy to pay, they will definitely try.
this is called "highway robbery" or sometimes "capitalism".

> The current settlements regime has evolved over many
> years of operational and business experience. ...

the current settlement regime locks new providers out of the
market, to the detriment of end-users everywhere.  go ahead,
tell me how you'd spend $1B to join the DFZ with pure SFP in
10 years, and then tell me how you'd ever make that money back
at the current/declining transit and on-net prices now seen.

but we digress.