Report on Peering and Transit Economics (etc)
2002-09-29 10:19:40
I have published an extremely detailed report on the economics,
technology, and politics of peering and transit. At
http://cookreport.com/11.08-09.shtml you will find the complete
introductory article, contents and list of 25 contributors to the
effort. The report centers on interviews with Bill Woodcock, an
essay / critique by Farooq Hussain and an edited version of the 6
weeks of discussion by the 25 contributors.
I include here the executive summary from the report. This summary
is not on my web site and I am not posting it elsewhere.
Executive Summary
Whither the Policy, Technology, and Economics of the Interconnection
of the Internet?
The collapse of the industry and of the price of bandwidth is
bringing significant changes into the ways in which ISPs and the
remnants of the Old Guard of Tier 1 backbones interconnect.
Some people who are affected have made some significant steps in
using NetFlow data in developing tools that are being refined into
what can function as bandwidth cost management systems. We identify
several explorations being taken in this direction and explore what
looks to be the most refined developed by Bill Woodcock with the
assistance of Alex Tudor at Agilent Labs.
Bill has developed a philosophy of interconnection that appears to
have a sound business model behind it. Bill's approach was
developed from the point of view of a small ISP that needs to
understand with as much precision as possible what it does cost to
get its bandwidth delivered. His model says that ISPs that are
multi-homed and have their own leased line customers need to peer as
much and as cheaply as possible. They also need to have two reliable
transit providers in case one fails. As long as their peering can
cut over to transit if it fails, he points out that economics would
seem to demand delivery of as much bandwidth by cheap peering as
possible to cut down on the requirement for expensive transit
bandwidth.
ISPs need to avoid local loop charges from their LECs and acquire
their own back haul to an exchange for inexpensive peering and if
possible a different exchange or exchanges for more reliable transit.
In order to figure how to most cost effectively architect their
networks they need to take and manipulate NetFlow samples of their
traffic in order to identify potential new peers via a study of the
traffic being delivered by their transit providers. If they have
automated tools to take samples from appropriate points, they can
over time get clear pictures of how their traffic is evolving through
actual NetFlow path analysis.
But Woodcock's colleagues seem to agree that he has done something
unique. He explains it in writing for the first time in this issue
of the COOK Report. Namely he does what he calls synthetic path
analysis by tacking his actual path data and doing a series of "what
if" transformations on that data. With the help of Alex Tudor from
Agilent labs he explains using actual data from January 31 2002 how
this synthetic analysis can be applied so that for the first time an
ISP, by plugging circuit cost data into its modeling software, can
know how much it really does cost to deliver its bits.
These ideas are new. While our experts agreed that perhaps 100 ISPs
may be doing some form of actual NetFlow data analysis, virtually no
one except Woodcock had done the synthetic path analysis. Avi
Freedman in his position as Chief Network Scientist at Akamai has had
ample occasion to use network routing and DNS to figure out data
flows. After studying Bill Woodcock's explanation found that he had
evidence from his own related experience that indicated Bill's
approach seemed valid. He points out that since 1999 he has been
doing a "what if" analysis on "Akamai flows" similar to Woodcock's
synthetic path analysis on router flows.
Our 50,000 word eight week long discussion involving 25 different
people contains a quite interesting dialog between the Avi and Bill
as they compare their approaches to the problem and conclude that the
ideas appear to be valid. However, we must also point out that Bill's
synthetic path analysis is not meant to be the sole criterion on
which to base peering and transit decisions. Once they have
identified potentially good peers, ISPs will find that factors of
geography and costs of interconnection at various exchanges may
become decisive factors in making their final decisions.
Although the largest carriers generally prohibit their technical
people from participating in this kind of discussion, we were
fortunate to get participation from large representatives of both the
cable modem and DLS worlds (namely Adelphia and SBC). At the most
general level these larger players seem to acknowledge the validity
of Woodcock's ideas. However, one things get specific, they maintain
that differences in the sizes of their networks prevent them from
placing too much faith in low cost cooperative peering. They seem
intent on joining and perhaps replacing the seven Tier 1 backbones.
As prices for transit bandwidth have crashed in about two years from
as much as $800 per megabit-per-second per month to in some cases
less than $100, the assumption that one should save transit costs by
peering as much as possible has become muddied. Our contributors had
a great deal to say about both the costs of transit bandwidth and
city-to-city OC pipes of various sizes. Carriers are extraordinarily
hungry for new lambda sales.
Farooq Hussain offers an essay that explains the international mess
that he believes peering and transit has become due to the Tier 1
oligopoly. Although the view one gets of traffic depends on the
places within the network where one measures, from some viewpoints
according to Adlex and other tools it would seem that while UUNET,
ATT, Level 3 and Sprint are holding firm in Tier 1, Genuity Qwest and
Cable and Wireless may be slipping seriously in overall traffic
rankings.
We reprint Bill Norton's comments on the death of ATM exchanges from
NANOG. Our contributors explain why the price differentials aren't
quite as crisp as Equinix would make them out to be.
Googin Interview on .13 Micron
We publish a mid May interview with Roxane Googin in which she
explains why the market arrival of .13 micron technology will bring
PC economics and costs to switches. Nortel and Lucent look unlikely
to rise and the LECs switching base will belong on the junk pile
sooner than anticipated. The Googinization of the LECs continues -
http://www.sbc.com/press_room/1,5932,31,00.html?query=20257. The
LECs of course are intent to keep the secret bottled up and whine for
regulatory relief that if delivered would only prolong the crash.
ICANN's Pathology
We summarize ICANN's latest egregious behavior from Vint Cerf's
trying to spin the loss in the Auerbach lawsuit to Joe Sim's temper
tantrum against Michael Froomkin. Froomkin has an outstanding new
law review article called Form and Substance and written in no small
part to lay bare Sim's obfuscations. From the predetermined award of
.org to ISOC - sorry Carl Malamud, you don't count in this new
commercial internet - to the regional routing registries public
rebuke of Cerf and Lynn ICANN's misbehavior marches on. The DoC
renewed its MoU this time for a single year. Bret Fausetts blog
grows better and Milton Mueller author of a new dispassionate study
of ICANN and the root finally throws up his hands in disgust on BWG.
One of the most amazing things about the ICANN pathology is how, in
the midst of the barrage of news about corporate malfeasance in
Enron, WorldCom, Qwest, Tyco and others, the collection of
figureheads who have agreed to serve as board members can continue
to go on on a daily basis in blissful assumed ignorance that they
have any responsibilities under California law.
We can only hope that Karl Auerbach who HAS taken his duties
seriously while his fellow Board members have shirked theirs, will
give some meaningful completion to his law suit by making it very
public and very clear how this group of people who have assumed they
are accountable to no one has in reality behaved.
RIAA goes Beserk
--
========================================================
The COOK Report on Internet, 431 Greenway Ave, Ewing, NJ 08618 USA (609)
882-2572 (phone & fax) cook(_at_)cookreport(_dot_)com Subscription info &
prices at http://cookreport.com/subscriptions.shtml Summary of
content for 10 years at http://cookreport.com/past_issues.shtml Info
on Economics of Peering, Transit & IXs
November - December 118 pages available at http://cookreport.com/11.08-09.shtml
========================================================
<Prev in Thread] |
Current Thread |
[Next in Thread>
|
- Report on Peering and Transit Economics (etc),
Gordon Cook <=
|
|
|