Hello,
I wonder if anyone has done any estimates on how many
percent of the Internet traffic:
A) does not pass a peering relationship
(i.e. is AS internal or passes through transit connections only)
B) passes a public peering relationship
C) passes a private peering relationship
I know this is fairly impossible to estimate globally, so even any
local or network specific research is appreciated.
Bon soir,
} On Fri, 5 May 2006, Aleksi Suhonen wrote:
} > A) does not pass a peering relationship
} > (i.e. is AS internal or passes through transit connections only)
} > B) passes a public peering relationship
} > C) passes a private peering relationship
Mikael Abrahamsson wrote:
} Judging from our perspective (TDC Song) my guesstimate is that we as a
} nordic ISP with majority of traffic to/from Swedish customers do:
} 50% traffic in Sweden
} 25% Nordic (Oslo/Copenhagen/Helsinki)
} 10-15% LINX/AMSIX
} 10-15% transit
I forgot to stress that I'm particularly interested in the ratio
between private and public peering.
Thanks for a swift response, by the way. 
The answer is, it depends. In any given region, prevailing market
conditions (the price of transit, etc), the costs of ix ports, the costs
of colo and fiber xconns, the geographic distribution of peers, and the
attitudes towards public and private peering are all going to have a huge
impact.
For example, public peering in Europe is FAR more pervasive than it is in
the US. Obvious reasons for this include:
European IX's US IX's
* Largely run by non-profits * Largely run by for-profits
* Largely un-associated with colos * Largely run by colo operators
* Early adopters of new tech like 10GE * Significantly lagging on new tech
* IX ports are generally very cheap * Same ports usually cost 3-8x more
* Larger numbers of smaller peers * Smaller numbers of larger peers
* Lots of language specific content * Lots of globally targetted content
Obviously market economics drive public peering much more in Europe than
in the US. To put it into perspective, the amount of traffic exchange by a
single "large" IX (such as AMS-IX) is roughly equal to all of the IX's in
the US combined. That same amount of traffic is roughly 1/2 (or less) of
the the traffic exchanged by a single "large" network (such as Cogent,
Level3, AT&T, ATDN, etc) via private peering.
Aleksi:
Too add some to the great Ras, for us, much has changed over the
years, as there has been more consolidation of facilities to be in,
globally, as well as changes in hardware, decreases in public and
private peering costs, cheaper fiber/dwdm, the end of atm, etc... So..
that being said, i would guess that many so-called large EU networks
would have ~ 5% of their traffic on public exchanges, and many of
those are not in USA exchanges or asian ones if their network goes
there. I would also guess that the greater, smaller networks are more
valuable to exchanges than the so-called larger networks from a
financial perspective, since they are probably sending /receiving much
more traffic.
Some exchanges publish their customer's mbs of connectivity and you
can draw your own conclusion from someone with 1000mbs vs. 10000Mbs of
peering capacity at an exchange. I think amsix is the one off the top
of my head that tells who/where/mbs to the switches. Lastly, there
are exchanges that allow for transit to be sold over them vs. not, and
that would greatly increase the possibility of getting a larger
network in there as well. Send me a private message if you want some
specifics as far as i can give them about 1299.
Peter Cohen