North American Network Operators Group

Date Prev | Date Next | Date Index | Thread Index | Author Index | Historical

Re: Peering vs SFI (was Re: Cogent/Level 3 depeering)

  • From: Paul Vixie
  • Date: Fri Oct 07 15:51:15 2005

[email protected] ("Schliesser, Benson") writes:

> > If it's still common for one to be billed only for "highest of in
> > vs. out" then there's no way to compare the benefits since there's
> > always a "shadow" direction and it won't be symmetric among flow
> > endpoints.
> 
> Thank you, Paul. I'd be interested in your feedback on these thoughts of
> mine below.
> 
> I do believe it is typical, perhaps with some variance but usually
> amounting to the same thing, that end-users are billed for the "highest
> of in vs. out" traffic, roughly the capacity they are provisioned. Thus
> if I may, I'll build on this to make a more concrete statement: each
> party in a peering relationship receives equal value for traffic
> exchanged. (traffic volume at the SFI translates into revenue from
> end-users)

value is subjective.  that's what's so funny about depeering announcements
and counterannouncements, where somebody always says words to effect of
"we're as big as they are, so there's no reason we should be paying them."
that MAY be true.  but it's not debatable.  either the former peer will see
it that way, or they won't.  value is subjective, not democratic.

so, i disagree that the parties receive equal value in the general case you
cite.  maybe they do, maybe they don't.  that's for them to decide.  each
of them, that is, to decide.

> Things aren't so simple in reality, though: you have to look at the
> element left out of my statement above, the "cost" of traffic exchanged.
> If one peer terminates more traffic than it originates, and the
> originating peer is performing "hot-potato" routing, then the terminating
> peer typically has a higher cost burden as it has to transport the
> traffic the greater distance. However the opposite holds true if the
> originating peer is performing "cold-potato" routing.

that summary is probably going to match the facts most of the time.  but i
know that cold-potato is really cheap for some people and really expensive
for others, and that the circumstances that really will govern the costs
and benefits of traffic exchange go way beyond how hot the potatoes are.

> Thus, such things exist as traffic in/out ratios between peers. But this
> is a blunt tool which seems to help enforce the exclusivity of the
> Tier-1 club, and actually acts as a barrier to competition.

it's only a barrier to competition if it restricts customer choice or
increases customer price.  in other words, person X doesn't get to accuse a
company of "restricting competition" because they're making it hard for
person X to enter the business.  person X would have to show that their own
choices, as a customer, had been constrained or their costs had been
increased in order to claim "barrier to competition."

otherwise it's just whinage.

the game is, you enter the field, you invest in some infrastructure and
build some channels, you pay what you have to pay to get access to the
network that existed before you, you charge what the market will bear, you
make up the difference (if there is one) with cash, and eventually you get
big enough to have enough negotiating power that your peering/transit costs
go down to where you have some margin.  if you run out of cash, you go
bankrupt and try again.  if you win, then you become part of the backdrop
against which new competitors enter the field.  if you charge so much for
access that new competitors need way more cash to get started, then you
also increase "what the market will bear" to the point where you're very
nearly helping new competitors as much as you're hurting them.  the only
way to win long-term is by staying efficient, never slacking, and being
frugal, especially regarding debt.

this leads to constant churn, even chaos and carnage, but ultimately it's
better for the customers, in my opinion, than regulated monopoly would be.

the thing that's irritating me at the moment is that network transit isn't
matching other technology curves.  as with the workstation on my desk, i
do not want to pay less for my new machine than i paid two years ago, in
fact i'm willing to pay the same, even adjusting for inflation.  what i 
want is to get a lot more for that price now than i got two years ago.
in transit pricing, that would mean we'd all be paying the same total every
month that we paid when prices were $1000/Mbit/month, but we'd have 20X the
bandwidth available.  instead of fighting over a stagnant market we'd be
competing to see who could get to the next order of magnitude faster (all
without invalidating our physical plant faster than we could depreciate it.)

but i digress.

> That is, anybody with a different traffic pattern (i.e., because of a
> different business model) will be excluded from the club despite the fact
> that they bring equal value in the form of traffic volume to the
> relationship.

the value they bring is only equal if the folks they're bringing it to say
it's equal.  see above.

> And club-outsiders are subject to increased relative operating costs
> (cost of revenue) compared to club-insiders.

there's a pain band.  if the other players charge you too little or too
much then they end up hurting themselves.  discovering that band and
tracking it as it changes is part of "staying efficient and never slacking."

> So what is the solution? "Warm-potato" routing seems possible
> technically, providing an approximation of cost-burden fairness.  Is the
> benefit worth the complexity to manage in practice?  And clearly, I'm not
> advocating endless open peering--the revenue element of the equation
> (customers) must exist. So what is the best way to determine the criteria
> by which a network is determined to be a "peer"?

for my part, i like aggressive peering, because i like high path splay, both
for end users and as a property of the system.  others here say that high
path splay means high route churn and artificially inflates running costs
including router CPU, router RAM, number of backbone engineers needed, number
of additional failures due to extra router CPU and RAM, number of additional
mistakes made by the additional backbone engineers, and so on.  YMMV.  warm
potato also appeals to me, but that's not something anyone can really enforce
on their peers.  as long as the internet remains a network of networks, there
will be an uneasy balance between unilateralism and multilateralism.  (ain't
it great?)
-- 
Paul Vixie