October 98 COOK Report published - public summary
Gordon Cook
DISPUTE WEB FARM BASED BUSINESS MODEL IS WEAK & MAY LACK INCENTIVE TO
ENGINEER TCP CONGESTION -- JOHN CURRAN DISCUSSES BEST EXIT & PAID
PEERING pp. 1- 11
Three years ago Exodus was one of the second tier national backbones
with peering agreements at the public exchanges. Today it is a
publicly held web farm company sitting on a war chest of a March 98
$69 million dollar stock issue and July $200 million dollar bond
issue. Led by the venerable Ellen Hancock of IBM and Apple fame,
Exodus, with only a temporary peering agreement with BBN, has become
a national web farm hosting such huge sites as USA Today and
GeoCities.
On July 9 Exodus was informed by BBN of a non renewal of its peering.
Instead of quietly renegotiating connectivity behind the scenes,
Hancock has taken Exodus' dispute with BBN public -- warning her
customers in a letter of August 5th that BBN and not Exodus would
bear the most pain as a result of the dispute. From what we can
determine after an exhaustive examination, it is Exodus that has
erred in calling attention to the weakness of its business model. The
Exodus model assumes that, since it is now a highly desired content
provider, it should be allowed to maintain its no cost peering
agreements with global backbones like BBN despite, according to some
estimates, an imbalance of traffic with BBN that is something on the
order of sixteen to one.
For the past two weeks BBN has been demonized on the NANOG mail list
by Exodus supporters and those who insist that Exodus get free
connectivity because it has content that BBN customers want. Exodus
reacted as though BBN were the only backbone that was breaking
peering with it claiming to have transit free interconnectivity with
all other providers. We conclude that at best Exodus may have cost
free peering with all other majors. But, even if this were true,
Exodus has absolutely no guarantee that it will continue past the
expiration of its current peering agreements. In reality we suspect
a considerable fudge factor. Certainly a reading of Exodus January
1998 S1 on Edgar makes it clear that Exodus was having to pay for
some of its connections. The caveats listed by Exodus in its
business model in the January 98 S1 are extensive and make for
instructive reading in the context of current round of finger
pointing.
We publish an argument by John Levine that, based on the assumption
that content is now king, justifies Exodus' position. We also
publish a discussion with Michael Dillion and Sean Doran where Sean
points out the very heavy burden maladjusted TCP web traffic can
impose on BBN's backbone performance. Of course if Exodus has the
right to transmit its content at no cost, it also will lack the
motivation to carefully engineer the behavior of its TCP congestion
windows with BBN.
We conclude our article with a 5,000 word interview with John Curran
done last Friday August 21. While John was careful to say nothing
explicit about any specific BBN peering situation, he offers the most
detailed description of both the industry's and BBN's approach to
peering that we have seen.
The principles that he explains are quite basic. First that peering
in the Internet has been and remains based on the assumption of
symmetry in traffic streams. In the case of BBN there must be a
ratio of traffic between peers that is not greater than 2 to 1. As a
result BBN has upwards of 50 peers. Second, the assumption in the
industry has been that both sender and receiver pay for traffic with
each paying about half the cost of meeting in what Einar Stefferud
recently referred to as the mythical middle. The problem arises that
the mythical middle vanishes when peering with a web farm causes
asymmetrical traffic. If BBN has to open up half a T3 of inbound
bandwidth for every T1 that Exodus has to install, the burden on BBN
becomes especially great when hot potato routing means that it goes
immediately onto BBN's backbone. If this did not happen and it were
Exodus responsibility to carry the traffic flow to an exit point
closest the location of BBN's customers, this could be a way of
restoring the symmetry destroyed by the web traffic. [The letter from
BBN leaked on the NACNet web site made it clear that BBN and Exodus
had been experimenting with this approach known variously as best
exit or longest exit routing.]
What is only now being recognized is that free peering relationships
with carrier backbones and large web farms engender traffic flows
that are so asymmetrical that we move essentially to a receiver pays
pricing model for the Internet. To quote Curran: "if we enter a world
where the senders don't really pay any incremental costs, you face
some huge implications. You end up with a situation where, for
example, a sender could decide to send you a large video image when
you connect up to his web site. . . And I guess I am just a little
bit concerned, if not from a business perspective, from a public
policy perspective about the multimedia spam possibilities when
senders are not paying any real incremental costs for sending more
information." Curran suggests that the industry continue to
experiment with best exit routing and with settlements based peering
where he defines that the content provider would pay only for the
portion of its traffic that fell outside of specified symmetrical
boundaries.
IPV6 TO BEGIN DEPLOYMENT NEXT YEAR --LIMITATIONS OF NAT TECHNOLOGY IN
IPSEC, IP TELEPHONY AND MOBILE COMPUTING TO FORCE CORPORATE MOVES
NUMBER ALLOCATION IS HIERARCHICAL WITH OPERATIONAL IMPROVEMENTS, pp.
12 - 16
In an interview with Jim Bound, who is the IPv6 technical leader for
the Compaq (formerly Digital) Unix group, we have created a very
broad and detailed overview of the status and functionality of Iv6.
As Bound explains it most of the protocol work is now done, a world
wide tunneling application known as the 6Bone is in operation and
Unix applications will appear next year along with initial moves to
allocate IPv6 addresses. He points out that NAT address translation
is a band aide approach that makes the use of IPsec impossible as
well as the kind of QoS applications needed for the success of the
internet telephony gateway market. Mobile roaming from laptops
back into corporate nets also won't work with NAT.
He believes that the urgency for corporations to use these
applications will compel them to start to implement IPv6 in their
networks next year. While they will have to renumber in
transitioning to IPv6, it has been designed to make the process easy
by comparison to IPv4. IP numbering allocation has been established
in such a way that, by means of top level, middle and local
agreggators, routing will be simplified and flaps unable to spill
over their aggregated boundaries. Unfortunately the vast number of
IPv6 address numbers available will not impact the hierarchical
delegation of the aggregators and we may see disputes over number
assignments continue.
BBN'S BLUMENTHAL DESCRIBES 1995 - 96 REGIONAL NETWORK INTEGRATION
PROCESS -- EXPLAINS DETAILS OF CURRENT OC-192 SONET BUILDOUT, pp. 17
- 21
We interview Steve Blumenthal, Senior Vice President of Network
Engineering at GTE Internetworking. He explains the steps that BBN
undertook to integrate BARRnet and SURANet with its NEARnet core
between 1995 and 1996. Critical to this process was a need to
standardize a myriad of systems from customer care and billing, to
network operations, maintenance and provisioning across multiple
organizations while maintaining the rapid growth of all three.
Lessons learned here are clearly relevant to MindSpring, Verio,
WorldCom and others acquiring existing ISPs. Blumenthal also talks
about GTEI's provisioning of its 24 strands of Qwest fiber. GTE is
using Nortel OC 192 SONET opto-electronics to light up the fiber.
The Nortel equipment supports 16 DWDM lambdas on each strand. GTEI is
beginning to provision OC3 and OC12 circuits from the fiber and are
offering ATM and frame relay services as well.
In order to provision high bandwidth circuits for sale to other
carriers GTEI has installed SONET throughout. Blumenthal covers
other aspects of GTE's planned expenditures of five to six hundred
million dollars a year for network upgrades for the next several
years. He concludes with and explanation of GETI's provisioning of
America Online and its global provisioning capability as the result
of its alliance with Equant, the commercial arm of SITA.
DOES IANA DO AN OPEN
PROCESS?, pp. 21-22, 24
Jon Postel and almost half the country code TLD administrators got
together in early August and tried to create among themselves a group
that would become the Names Council and speak for the commercial
registries as well. We show that primarily behind this plan are ISOC
and the country code administrators who, while charging lucrative
fees, pledge to keep their organization beholden to Jon Postel and to
support his IANA draft by laws. We list the fees charged by those
countries requiring no operational presence for the administration of
their domains.
We also note the conclusion of Roeland Meyer on August 22 was even
more striking: "Actually, you make a real strong point. If the IANA
folk have a discussion list, it is certainly closed to me, as well
as, most that I know. It could be argued, by Postel and Co, that they
don't have time for such silliness. However, he's not available to
anyone else I know either. If he's doing something, and I'm sure he
is, why do we all find out after the fact? That kind of blows his
credibility vis-a-vis open processes. On reviewing the latest Draft-
Postel, this is kind of underlined (during my feedback to the NSI
review, which Dan and I both did). Jon Postel doesn't "do" open
processes."
***************************************************************************
The COOK Report on Internet New Special Report: Building Internet
431 Greenway Ave, Ewing, NJ 08618 USA Infrastructure ($395) available. See
(609) 882-2572 (phone & fax) http://www.cookreport.com/building.html
co...@cookreport.com Index to 6 years of COOK Report, how to
subscribe, exec summaries, special reports, gloss at http://www.cookreport.com
***************************************************************************
Tim Salo
> From: Gordon Cook <co...@cookreport.com>
> Subject: October 98 COOK Report published - public summary
> [...]
> DOES IANA DO AN OPEN
> PROCESS?, pp. 21-22, 24
> ... Jon Postel doesn't "do" open processes."
Joh Postel is scheduled to talk about IANA.ORG at 7:30 this even at the
IETF.
Will you be there Gordon?
-tjs
Randy Bush
> ... Jon Postel doesn't "do" open processes."
Gordon, get a clue. Jon Postel was doing open net processes before Gordon
"reality distortion press" Cook discovered the typewriter, let alone the
computer. Jon is world-reknowned for this.
randy
Russell Nelson
I must interject a note of reality here. A while back Jon sent out a
note to the .us domain delegatees, wherein he proposed charging for
having a domain name delegated to you. I thought that deserved
discussion, so I started the us-domain mailing list (since disbanded)
in June of '97. Jon joined the mailing list, *did not participate*,
and after a period of time, pronounced a change in the .us domain
policy which was unlike any proposed on the mailing list, and whose
implementation was unaffected by any following discussion. *You* are
welcome to call this an open process. I do not.
--
-russ nelson <rn-...@crynwr.com> http://crynwr.com/~nelson
Crynwr supports Open Source(tm) Software| PGPok | Freedom is the primary
521 Pleasant Valley Rd. | +1 315 268 1925 voice | cause of Peace, Love,
Potsdam, NY 13676-3213 | +1 315 268 9201 FAX | Truth and Justice.