4/98 CIMI Netwatcher Management Briefing. Info on VPN proposal by Savvis
[Isn't Savvis an ASND customer]
datacomm-us.com
A manager can't move these days without running into a comment about the Internet. Everybody knows it's
changed attitudes about TCP/IP, about electronic advertising and commerce, and about public data networking
in general. Most don't know much about how it works, either in a technical sense or as a business. That's risky,
because something that has had as much impact as the Internet in the past just might be important in an
ongoing sense.
The Internet is an outgrowth of something started by the Department of Defense in the late 1960s. DoD had a
number of universities and Federal research agencies working on various hush-hush projects, and needed a
way for them to communicate data among themselves. To provide it, the Defense Advanced Research Projects
Administration (DARPA) funded the creation of a computer network called ARPANET.
The original ARPANET didn't even run TCP/IP; it didn't exist at the time. That protocol was developed in the
late 1970s, and was phased into ARPANET through about 1980.
Expanded use of the network led, in the middle 1980s, to a shift of funding from DARPA to the National
Science Foundation. NSFNET, the successor to ARPANET, was based on T1 backbone trunks, transitioned in
the 1990 timeframe to T3.
By the early 1990s, the "Internet" was still primarily a collection of research and university organizations, but
an increasing number of corporations were added through connections via one of the core network members.
The commercial interest in the Internet surged with the introduction of the Worldwide Web, and through the
early 1990s NSF funding phased out and the Internet became truly commercial.
The commercial Internet is made up of Internet Service Providers (ISPs). These providers offer leased-line
attachment, dial-up, or both, and serve not only the business community but the growing number of residential
users. Online service providers like AOL, who offer connection to the Internet, are essentially ISPs who have
services that encapsulate and augment basic Internet services.
Within an ISP domain, the ISP supports a protocol (TCP/IP) and provides directory services to permit locating
each of the permanent content resources the ISP connects. Information consumers can access content within
their own ISP without touching anything that we'd strictly call the "Internet".
To provide information consumers access to content on another ISP, the ISP networks have to be
interconnected. This was first done through a series of Network Access Points (NAPs, pronounced like the
nap you take when you're tired). NAPs are interconnection facilities run under contract by a service provider
and available to all. ISPs run a line to a NAP and they're connected to everyone else there.
NAPs, needless to say, tend to be points of congestion. A second level of interconnection was added to help
this problem: the Metropolitan-Area Exchange (MAE, pronounced "may"). MAEs link service providers in a
region and provide a faster way to transfer information between local ISPs there.
For most providers, even the combination of NAPs and MAEs were too slow, and the service providers began
to develop strategies to link their own networks directly - something called "peering". A peering agreement is
a negotiated relationship between ISPs that provides for network interconnection under a given set of cost and
technology conditions. Networks that are peer-connected have much faster mutual transfer capability, and so
peering is sought by providers who want to either have access to content on another network, or who want to
provide their own content to that other network's consumers.
Technically, all the interconnects are built either as LAN switches or ATM switches, linking border routers
running the BGP4 protocol. This facilitates control of the relationship and prevents "hacking" of address tables
and other destructive behavior.
The peering agreements, and the underlying interconnect problem they represent, are a good place to start
considering the business issues of the Internet. Common carriers have interconnect points, and there are
usually treaties developed to cover how compensation will flow, etc. So far, this has eluded the ISP world.
Each ISP collects money from its customers. In general, the money is a fixed price per month for "access", and
thus includes whatever traffic the user can push through the access pipe. It is rare for there to be any form of
usage charging.
There are two problems with this model. The first is the uneven revenue potential of various customer types,
and the second is the impact of Internet businesses on the carriers.
Consumers, meaning residential users and businesses who use dial-up access, are essential to the electronic
marketing mission that has really justified the Internet. Let's face it, people don't publish on the Internet to
educate, they publish to sell. That means that we have to get those we want to sell to a presence on the
Internet as well. Cross-subsidization of consumers by providers is logical in such a scenario.
It doesn't always work out on the Internet, because "consumers" are often owned by one ISP and providers by
another. Many ISPs lose money on dial-up users (we're told the average loss is $10 per month per port, or
about a buck per user). The ISPs who don't do dial-up are immune to this loss, of course.
Some revenue relief can come to the "retail consumer" ISPs from advertising, but it's still true that the ISPs who
serve business users are the most profitable. That means it may, over the long haul, become more difficult to
support residential users or dial-up in general. This problem could put pressure on the desire to expand the
consumer population on the Internet, or even result in service degradation to those already on.
The second problem of Internet business is also a potential degrader of performance. There are a lot of
companies who claim to be saving millions with the Internet, in product delivery and customer support. That
"savings" is translating to a business problem for the ISPs, because every new traffic source on the Internet
creates infrastructure pressure on the ISPs. Since traffic isn't charged for, there's no revenue source to offset
the cost, and the ISP must either upgrade without compensating revenue, or let the network's performance
deteriorate.
The disconnect in the ISP business space is also a potential problem for new revenue-generating services like
VPNs. Virtual private data networks could earn ISPs as much as $10 billion by the year 2000, but these services
require explicit QoS levels that the current router-based Internet can't provide. Some ISPs with ATM
backbones can provide VPNs within their own network span, but VPNs can't readily cross ISP boundaries
unless the interconnect points are capable of maintaining QoS. There are no consistent standards for QoS
across an interconnect today, and there is no precedent for how VPN revenues would be shared by two ISPs
who cooperated to provision the services.
Some hope exists for this situation, but not a lot. The Internet community, bred from university counter-culture
people opposed to any form of regulation, would surely resist (as they already have) any kind of governmental
intervention to clean up the mess. Given this, there is an undercurrent of non-competitive activity developing.
Big players like MCI, UUNET, or Sprint have enough network mass to be credible in the VPN market without
partnership. If these players can tap the multi-billion-dollar VPN revenue opportunity and others cannot, the
large ISPs will quickly eat everybody else and the broad and innovative nature of Internet carrier services will
disappear.
Some recent steps to solve this problem have been proposed by Savvis, a St. Louis ISP who is also
Boardwatch's number one ISP pick. Savvis proposes a "Brokered Private Peering" that would establish
multi-ISP participation in a consortium aimed both at formalizing and opening the peering process and
expanding peering to provide for QoS-based services in both a technical and financial sense. Such a
consortium could bring scale and mass to the smaller ISP players, making them VPN equals to the big guys,
and eliminating the risk that the whole ISP community would collapse into a group of carriers who were much
the same players as existed in the common carrier space.
It's too early to say how this venture will progress, but it is fair to say that the next five years will bring about
major changes in the Internet-one way or the other.
In the Know
In our last issue, we developed an overview of the evolving MPLS standard. This month, in an expanded In the
Know that will displace the Strategies topic as well, we'll take a deeper look at the concepts of MPLS. This
topic, in the current issue and in subsequent issues, is for subscribers only; it will not be posted to our web
site.
This material will expand on the summary of MPLS we've already provided, and explore issues like MPLS'
architecture, its relationship with VPNs, and its implications on the service DMARK.
NETWATCHER Index Page
Access the index of CIMI Corporation's recent newsletters.
DataComm-US Home Page
Explore the Web's resource site for the data communications industry.
April 1998 Volume 16.4
Netwatcher (ISSN 0890-5800) is a monthly publication of CIMI Corporation. Subscription information is
available here . Copyright c 1998, CIMI Corporation. All rights reserved. No publication or reproduction of this
document is permitted without the express written consent of CIMI Corporation. |
