So how do you reconcile the increasing number of private facilities? Although
not exchange points in the truest sense ala MAEs and NAPs, they do carry an
increasing percentage of cross-network traffic. The number of these types of
facilities will grow. Maybe our difference here is that I count these and
others do not. des
My point was not in relation to the NAP vs Private interconnect debate,
nor the many NAP vs few NAPs debate, just when measuring "few" or "many"
they should be done in the context of telecoms pricing and locality
of content, and thus "NAPs per country" not "NAPs per continent" was the
appropriate ratio. I've since been told telecoms pricing between
Canada and US is almost as bad as between (say) UK and FR in terms
of the hike as you go over the border. I don't think many people would
argue each extra US NAP diminishes the case for an extra Canadian NAP.
However each successful US west coast NAP does probably diminish the
case for another successful NAP on the US west coast, or possibly the US
in total.
FYI my view on the private interconnect versus public IXP interconnect
(probably not very informed as I don't have many private interconnects
with Sprint et al 
) is that if public IXPs worked as well as private
interconnects (didn't get overloaded, no interference to traffic from
third parties, peer to peer bandwidth always available at STM-1 or
above etc. etc.) there would not be nearly so much of a case for private
interconnects (the "people whinging about peering" argument is valid
but not relevant here as I'm assuming tier 1s would always connect
to at least 1 IXP anyway and whinging is not proportional to no. of
exchange points connected to I guess). In fact one could argue that
we already have "public" IXPs run like this - the SDH/SONET equipment that
runs the private circuits on private interconnects. Other than their
invisibility to the customer, and the protocol, these aren't doing
a very substantially different job from an ATM switch with VCs
between different (off-site) peers. Perhaps someone should persuade
Cisco to make a card running channelised STM-15 and start an IXP
running with just mux/demux kit. Or maybe not...
More seriously the argument about routers and bottlenecks is
relevant too - see Vadim's paper.
Alex Bligh
Xara Networks