Equality and asymetry in network designs

Tue Aug 25 03:14:39 UTC 1998

jcurran at bbnplanet.COM (John Curran) writes:
>p.s.  The fact that the sender of traffic should be paying some portion
>      of the resulting costs is not a surprise to anyone; many of the
>      content companies that I've spoken to believe they already are
>      paying more as traffic increases, and were quite surprised to 
>      find that it doesn't actually make it to the networks which 
>      bear the brunt of the traffic carriage.

As someone who pays to carry traffic to both coasts, and pays to carry
traffic from both coasts to the center of the country and other places
where I notice very few large providers seem to exchange traffic I am
perpetually amused by this recurring discussion.

If I read Mr. Curran's statement at face value I would assume networks
such as DRANET which are net importers of data, and pick up the data on
both coasts networks would expect to start receiving payments from large
providers with large web hosting centers in cities such as Cambridge.

I look at maps such as http://www.bbnplanet.com/products/maps/us_host.htm
and see a network which is strong in some areas, and very weak in other
areas.  St. Louis seems to be a gaping hole in many large providers.  Sure,
the sales people always say "don't worry, any day we'll have capacity there
soon" but if you want something today we have to pay our way to an outer edge
of the country.  Even if the network shows a "dot" in the middle of the
country, when you investigate further, I've usually found limited facilities.

I'm sorry if I sound cynical.  But I've been down this road of ever
changing requirements too many times.  "Best-Exit" used to be the norm,
I even have a mail message from a Sprint engineer prohibiting "Hot potato"
routing if I wanted to peer with Sprint.  Then the requirements changed,
I modified our network to comply, and the other provider of interest comes
up with a different requirement to justify it isn't a "peer" because something
is "unequal."  The fact is, no two networks are equal in every way.

How do you compare networks with large consumer access, web hosting, or
large numbers of dedicated access customers?  How do you compare a network
with a large backbone in the continental USA with a network with a smaller
backbone in the USA plus Canada?  Or how to you compare a network with
lots of transatlantic connectivity, but little cross-USA backbone or
trans-pacific connectivity.

And the biggest question of all.  The USA is still the largest net exporter
of data bits in the world.  How much should US providers be paying non-USA
providers to accept the bits?  Right now, other than some NSF-funded
connectivity such as Sprint ICM, most non-USA commercial networks must
pay the full cost of the inter-continental circuits.  Should the Internet
adopt a charging model like international voice settlements. Will we see
the rise of 'call-back' services on the Internet based on asymmetric charges
for inbound and outbound data flows?  The simple way to avoid people
'working' the system is to keep charges as close as possible to the actual
costs, so there is no incentive.  But shareholders prefer companies that
operate well above the marginal cost, so you see companies trying to
create a pricing model which allows them to charge above cost.

I like profit too, but other than certain regulated monopolies and software
companies, in a competitive market with low barriers to entry profit
margins are going to be narrow.  One way to increase the 'spread' is to
to raise your competitors costs and make it difficult for new providers
to enter the market.  Internal network design requirements of the other
providers network seem to be the most popular, e.g. you must put pipes
of certain sizes into certain locations even if it doesn't make design
sense.

I've noticed that most of the newest competitors trying to enter the
market in the last year have resorted to buying a provider with an
existing peering agreement because it was virtually impossible for the
new competitor to get peering with some large providers, Level 3->Geonet,
Qwest->Eunet, and then pumping up the size of the pipes through those
existing interconnect.

I've long said I'll agree to any settlement scheme, so long as it is
reciprocal.  The problem I've seen with most of the schemes going back
to the ANS CO+RE charges has been the one-way nature of the settlements.
-- 
Sean Donelan, Data Research Associates, Inc, St. Louis, MO
  Affiliation given for identification not representation