Notes on the Internet Transition
This is a copy of an email message assessing the impact of the transition
of the Internet away from an NSF-subsidized NSFNet
backbone. It was sent to the Unidata community email
list by Dave Fulker in late September. Ongoing inquiries
indicate that this assessment is still accurate.
Dear Community Members,
Stimulated by John Horel's recent message on the "Demise of NSFNET,"
this note is intended to provide some degree of clarification on the
transition presently underway in the operation of the Internet. Of
particular note is the elimination of the current NSFnet backbone as
the means for interconnecting regional networks.
NAP stands for Network Access Point. Four NAPs (near Chicago, New
York City, San Francisco, and Washington, DC) are being funded by NSF
to replace the functionality of the current NSFnet backbone by
providing interconnections among Network Service Providers (NSPs)
which, in turn, interface directly to the regional networks. Under
this model, the end-to-end connection between two universities, U1 and
U2, which are connected to distinct regional networks, R1 and R2,
would be of the form
U1 <--> R1 <--> NSP1 <--> NAP <--> NSP2 <--> R2 <--> U2
where NSP1 and NSP2 are the Network Service Providers used by the
regionals, and where NAP is the Network Access Point best suited for
this particular NSP1-NSP2 rendezvous. Of course if NSP1 and NSP2 are
the same, the NAP drops out of the picture, just as current backbone
plays no role in communications between universities served by the
same regional network.
Thus, the NSP1<-->NAP<-->NSP2 portion of the diagram provides exactly
the same functions as the current backbone, built upon the aggregate
capabilities of the NSF-funded NAPs and the privately funded NSPs. As
I understand it, NSF initially will help each regional network pay one
of the NSPs for T3-equivalent access. However, the regionals are free
to use the available NSP services in other ways, such as Westnet's
multiple T1 configuration referenced in John's note. Also, this NSF
subsidy will be phased out in a few years.
As I perceive it, we are entering an era of uncertainty and change,
but not of radical change. It appears to me that John's situation is
at least somewhat unusual, as indicated by the material which follows.
In contrast, most universities will see little change in service,
though their networking costs certainly will rise as the subsidy
decreases. We have been anticipating increases and have believed
their magnitude would be on the order of 10%--a number I can neither
correct nor confirm at this stage.
Supporting this view is a recent note from Dan Vietor (15 Sep) in
which he relates comments from Scott Ballew, network administrator for
Purdue. According to Dan, "The conclusion Scott makes is that our
network connections aren't really changing here in the short run and
we shouldn't have any problems with the new network connections even
though they won't be the same as in the past."
With this picture in mind, you may or may not want to read the rest of
this lengthy message, composed primarily of excerpts from other
messages.
The first excerpted message arose because I forwarded John Horel's
original message to the Reference Desk at InterNIC. Established with
NSF funding in 1993 to make networking and networked information more
easily accessible to researchers, educators, and the general public,
the term InterNIC signifies cooperation among three Network
Information Centers or NICs, one at AT&T (Directory and Database
Services), one at General Atomics (Information Services), and at
Network Solutions, Inc. (Registration Services) all run as a part of
NSF's role re the Internet.
Kent England, Executive Director of the InterNIC project, sent me an
informative response which is included below (denoted by ":" in the
first column). John's original text is denoted by ">", and other
annotations are my own. There are various references to CERFnet,
which is one of several regional networks; I think most of CERFnet's
clients are in southern California.
:Subject: Re: Transition Concerns
:From: kwe@cerf.net (Kent W. England)
:To: Dave Fulker
:Cc: refdesk@internic.net, susan calcari
:Date: Mon, 19 Sep 1994 15:32:02 -0800
:Message-Id: <199409192228.PAA15588@is.internic.net>
:
:Dave;
:
:I was today forwarded your question to the Reference Desk. As the Exec
:Director of CERFnet and the InterNIC project I feel you deserve a personal
:answer from me to your questions.
:
:The answers across the Internet about what happens during and after the
:transition vary greatly. I have asked my associates at SDSC and my
:technical staff to tell me more about the NCAR situation, but you should
:probably talk to the NCAR networking folks directly to determine what the
:details are regarding their situation.
:
:Let me give you some brief and general answers to the posed questions:
:
:>1) At what speed will the NCAR computer facility be connected to the
:> internet in the near future? Ditto for Unidata and other entities in
:> Boulder?
:
:I don't know the specifics of NCAR, but will try to find out.
:>
:>2) Is $30000 the going rate for a T-1 line and what factors affect the cost?
:
:An Internet connection is composed of several services:
:
:1) a leased line connection from a premise to the point of presence of the
:Internet service provider. This could include mileage charges and so could
:vary considerably. A T1 could cost $1000 per month depending on mileage.
:
:2) equipment. Specifically a router, CSU/DSU, dial-modem, UPS, etc. This
:is valued at about $5000 or more. T3 costs much more. Charge of $500 per
:month or less.
:
:3) Internet service fee. This is what the Internet access service provider
:charges for service. Probably in the range of $1000 per month or more.
:
:So, a T1 Internet access service could cost $2500 per month or $30k
:annually. It could also cost less or more.
:
:>
:>3) Will any university have access to the internet at speeds higher than T-1
:> after the first of the year?
:
:Yes. For example, there are several University of California campuses on
:CERFnet as backbone sites. They have Ethernet access to our T1 and T3
:backbone, so they have >T1 peak access to the Internet via CERFnet. There
:are other universities on other regional networks that have Ethernet or
:FDDI access to the regional backbone. MIT on NEARnet for example. The
:typical university connection to their regional network is unchanged after
:the transition, unless that connection was a connection directly to the
:NSFnet backbone and does not support a supercomputer center.
I believe the University of Utah falls into this special category,
being attached directly to the current (T3) NSFnet backbone but not
being considered an NSF supercomputer center.
:
:>
:>4) If your institution relies on T-1 links, how many T-1 connections do you
:> have access to and how many institutions rely on those links?
:>
:>5) According to the information provided to us, there are no viable
:> commercial links (Sprint, ATT, etc.) at speeds higher than T-1. Is this
:> true?
:>
:T3 service will be available from several providers who are going to
:provide connectivity to the NAPs for the regional networks. I don't know
:whether all services will be offered in all locations and these are new
:services for these providers. These are very expensive, but amortized over
:many regional customers, they are affordable and will be in part subsidized
:by NSF grants to the regionals.
:
:CERFnet will be getting T3 service for our transition.
Through separate communication from Pat Burns, Director, confirming
John's information that Westnet (the regional network which serves the
University of Utah) has chosen to get multiple T1 circuits instead,
based on a number of technical and economic factors.
:
:I hope this helps. My understanding is that most regionals will be getting
:grants to cover the cost of R&E traffic that the NSFnet used to carry.
:Each regional is responsible for providing for its own alternate service to
:the NAPs. The NSFnet will remain only to serve the supercomputer centers
:until the vBNS is built and operational.
R&E means research and engineering, and vBNS stands for the very
high-speed Backbone Network Service, intended primarily for connecting
supercomputer centers, including NCAR.
The following is a verbatum copy of the aforementioned message from
Scott Ballew to Dan Vietor in response, I believe, to Dan's forwarding
of the Horel note.
:Date: Thu, 15 Sep 1994 14:06:26 -0500
:From: Scott M. Ballew
:To: Daniel Vietor
:Subject: Re: NSFNet info?...
:
:First, some history. In the mid 1980's, NSF funded a project to link
:their supercomputer sites to each other. This was the first NSFNet
:backbone. A few years later, the DoD decided to get out of the general
:network backbone business and decided to shutdown the ARPANet (the
:backbone of the Internet). NSF proposed the expansion of the NSFNet to
:replace the ARPANet as the backbone, and this occurred in 1989.
:
:At that time, Purdue had a direct ARPANet link (at 56k, it was none too
:fast :-). As the ARPANet went away, we established a T1 connection to
:UIUC, an NSF supercomputer site, and thus on the NSFNet backbone. (After
:some problems) with our connection, (we) opened discussions with Advanced
:Networks and Services (ANS). ANS was then a not-for-profit company
:created by IBM, MCI, and Merit to manage the NSFNet backbone under
:contract to NSF. They were also entitled to establish direct subscriber
:connections. Shortly thereafter, they became a full-blown network service
:provider, and other providers began screaming about the potential conflict
:of interest in having ANS be a network service provider *and* a backbone
:service provider. NSF put the backbone service out for bid.
:
:The results changed little (ANS won the bid), but the seeds of the present
:day were planted. NSF finally decided they wanted out of the backbone
:network business, and announced that they were discontinuing the NSFNet.
:Their proposed replacement was a set of NSF sponsored Network Access
:Points (NAPs) where the Network Service Providers (NSPs) could
:interconnect with each other. They imposed a few policy constraints on
:the NAPs and NSPs to ensure a short-term (5 years, or so) guarantee that
:the US Research and Education community would continue to be able to
:communicate with each other. ANS is one of these NSPs, and is complying
:fully with the NSF policies (they must if they want to do business with an
:NSF funded site). Other NSPs are doing likewise. The regional networks
:must also subscribe to an NSP, or must connect directly to a NAP and
:negotiate their interconnects with the NSPs. Many sites who had
:previously "gone it alone" were faced with connecting to their regional
:networks, or of paying whatever the NSPs might ask. This is the
:transition to take place on Nov 1, 1994.
:
:Ok, with the history out of the way, how is Purdue positioned in all of
:this? Currently, we have a T1 link provided by ANS (which also happens to
:terminate at the Chicago NAP). This will not change. We pay ANS an
:annual subscription fee (vague memories place this at around $60-70k) and
:this is not slated to change. Since ANS is an NSF-approved NSP, they will
:provide full connectivity to all other NSF-approved NSPs, and to any other
:NSP who will negotiate with them. In short, our ability to communicate
:with any other site will not change because of any changes on our end.
:However, they may get a smaller pipe (or have to share a larger pipe) and
:that may impact how we communicate with them. This has yet to be seen and
:will have to be explored on a site-by-site basis.
:
:As a US R&E site, we are also well situated with respect to foreign sites.
:The NAP agreements include provision that all traffic that arrives at a
:NAP with a US R&E site as one endpoint, must be delivered, even if the
:other site is a foreign site. The problem most of the European and
:Pacific Rim sites face is that they used to be able to throw traffic for
:each other at the US backbone and we would fund its traversal of North
:America. No longer will this be the case. They will have to have transit
:agreements with one or more NSPs to get their traffic from the East Coast
:to the West Coast (or vice versa). This, needless to say, does not please
:them. :-)
:
:So, the long and short of all this is that nothing much should change in
:the short term (there may be some burps as it transitions over, but they
:should be resolved fairly quickly). In the long term, it may very well be
:that our network costs go up. What impact that has is still very much up
:in the air. If they rise high enough, it may be hard to convince the
:administration that our network connection is important. We shall have to
:see, but that has always been the case.
:
Hoping you find this information helpful,
Dave Fulker, Director, Unidata Program Center
P.S. For more information from the InterNIC, use Mosaic and open the URL
http://www.internic.net/infoguide.html
or use Gopher to connect to
is.internic.net/11/infoguide/
Within either of those, look under "Getting Connected to the Internet"
to find lists of access providers.
----
