mailing list archives
Re: Level 3 Peering Guidelines
From: Leo Bicknell <bicknell () ufp org>
Date: Fri, 26 Aug 2011 11:46:47 -0700
In a message written on Fri, Aug 19, 2011 at 05:40:43PM -0400, Patrick W. Gilmore wrote:
Yes, Above.Net broke the original peering-ratio fight that way. Thank you for that. Too bad it didn't last.
I forgot to write back when you first posted this, but the recent
follow ons reminded me...
While it is not a wide spread practice, I am 99.99% sure several
networks are still using this method in private. I also know of
at least two networks that have in the past deaggregated their large
supernets to peers to affect a similar result.
The interesting thing is that this is not done to elminate the
peering ratio argument in the aggregate, but to make up for the
geography of the two networks. Maybe one network insisted on peering
in a city that the other only has a few customers in, so the link
is mostly idle while others run too hot. Rather than provision
more bandwidth they agree to meds or deaggregates to get the traffic
to the lightly loaded link. The ratio in the aggregate is basically
I've said it before, more times than I can count. Peering is a
business tool, a means to an end. The goal, the 'end', of for-profit
companies is to make a profit. Sounds obvious, but surprising how
many people forget this. If peering with another network will
increase your profit and you turn down the peering request, you
should be fired. Your ego should not be substituted for business
I agree with your premise, and think everyone involved _thinks_
they are making the best business decision for their company.
However I believe many folks are in fact making poor decisions for
their company because they lack data. It's interesting to look at
the first order costs of peering; the port, cross connect, and time
to set it up. In my experience though the second and third order
effects are far more important; how it affects your ratio with other
peers, your customer's perception of your network performance, etc.
These are much harder to measure, and near as I can tell virtually
no network does.
I've been asked many times how the government could step in and
help, aka regulate peering to make it better. In thinking about
all the things that could be mandated, the interactions that can
be regulated, I see them generally as a univeral bad. With one,
major, glaring exception....
Right now peering is all done in secret. You don't really know if
your competitor is getting free peering, paying to peer, or getting
paid to peer. You don't know if they are really insisting on the
same locations with other parties they insist with for you. It's
a perfect game of poker. This is in stark contrast to say, transit,
where you can get prices from 20 people and compare.
AboveNet also used to put graphs of almost all of its peering links
online, in real time. A few peers flat out refused, but most were
there. You could see where they peered, at what speed, and how
much traffic was flowing. You could also calculate the ratio.
I think the Government could bring a lot of sunshine to the party
if it simply required all ISP's to do just that. Post a graph of
every peering link or port to a public exchange, updated in real
time. They all have the data internally, they are all collecting
it for their own needs. You still wouldn't know if any money is
changing hands, but imagine how it would change many of the
interactions. Oh, you require me to hit 3Tb aggregate, when you
have 20 peers that are less than half that? Really, you're going
to require me to peer with you in Nome Alaska when you don't have
any other peers at that location?
It would also let ISP's make an informed decision about depeering
events. The next time cogent splits from someone you can go look
at the graphs on both sides, and make your own decision who was
being more reasonable. It would also let you, the transit customer,
evaluate the extent and free capacity to peers your ISP has before
I have yet to find a down side to this sort of sunshine. I'd wecome
anyone who thinks its a bad idea to educate me.
Leo Bicknell - bicknell () ufp org - CCIE 3440
PGP keys at http://www.ufp.org/~bicknell/