[1805] in Commercialization & Privatization of the Internet
Re: Fair competition
daemon@ATHENA.MIT.EDU (Hans-Werner Braun)
Sun Dec 29 14:14:52 1991
From: Hans-Werner Braun <hwb@upeksa.sdsc.edu>
To: com-priv@psi.com
Date: Sun, 29 Dec 91 11:11:44 PST
In-Reply-To: <9112290040.AA26013@rodan.UU.NET>; from "Rick Adams" at Dec 28, 91 7:40 pm
I would like to clarify a few points. My objective if not to defend or
accuse, just to set some things straight.
>From: rick@uunet.uu.net (Rick Adams)
>
>In 1987, Merit et al won the right to lose money providing a DEDICATED
>network connecting the various regional networks. That event was fully
>and fairly pubilicly competed and I don't think anyone involved has
>ever said otherwise.
The sole winner of the award was Merit (a consortium created by Michigan
Universies to provide networking support; created in the late sixties).
Merit had joint study partners, IBM and MCI, and also got supplemental
funding by the State of Michigan. Certainly Merit is not loosing money.
Where would they get it from? The State of Michigan also contributed,
as it is interested in Michigan's technological evolution. Leaves IBM
and MCI. I don't think they were in to loose money. Gain can be measured
in many ways. Yes, both IBM and MCI contributed in-kind, in particular
equipment, support and expertise. Yes, it did cost them alot of money,
but they also gained. IBM, for example, did not have the networking
reputation in 1987 that they have today, except, of course, in a SNA
environment. I have heard that today more of their products support TCP/IP
SNA. They gained alot of insight into the Internet and are viewed much
more seriously in several environments, networking-wise. I am sure they
were able to sell things, backing it up with their NSFNET expertise. Same
with MCI, kind of, getting heavily into data networking. I don't think that
either IBM or MCI were planning for taking over large scale TCP/IP computer
networking support in 1987, or were even planning for a company like ANS.
That evolved over time. However, again, both IBM and MCI gained in a variety
of areas, not necessarely measured in funding flows related to the NSFNET
project. They took quite some risk, though, and should be cheered for that.
I quite remember all the flames in early 1988 that the undertaking was
impossible to accomplish.
I am only talking about T1/1987+ so far!
>Now, here's the key point. The so-called "T3 upgrade" isn't an upgrade.
Well, first of all the T3 support was already mentioned in the initial
proposal in 1987. And the upgrade was even based on the time figures
mentioned in 1987 as an extension of the T1 agreement. If you look at
the NSFNET solicitation from 1987, it did not even *specify* line speed.
That was a bit open to the proposer, and mentioning T3 earlyon (in the
1987 proposal) seemed strategically important enough to believe it would
help with the award process (in 1987).
>It is the dismantling of the T1 network and replacing it with
>SHARED SERVICE from a commercial provider.
It may appear like that from the outside, but it is not really correct. The
T3 upgrade was planned by the NSFNET partnership in pre-ANS days. It would
have happened with or without ANS. The technical implementation "dismantling
of the T1 network" was elected for pure pragmatic reasons. We reached a point
last year where people would have crammed bad things down our throat if we
would have disrupted the T1 NSFNET backbone services. So, the T3 network was
planned as an overlay over the T1 network to keep the T1 network alive until
the T3 network was proven to work to then phase the T1 network out. This is
an implementation detailed, was planned before ANS existed and ANS had nothing
to do with it. In fact, it is the same thing we did with the old 56kbps
NSFNET backbone in 1988 (creating an overlay, phasing things over, and then
dismantling the 56kbps network (I think some of the Fuzzballs (56kbps nodes)
are still in place for network time servers, actually)). ANS then happened
later in 1990, trying to do whatever they are trying to do, including enriching
the infrastructure by even more connectivity.
>My point of view is that I could have done a hell of a lot better job for
>$10 million.
Rick, what exactly would you provide if you had $10M? I am quite seriously
interested. Assuming, of course, you would want to provide a national T3
network with it.
>From: cook@tmn.com (Gordon Cook)
>
><<MESSAGE from>> Gordon Cook 28-DEC-91 21:57
> cook@tmn
> One thing i note about the T-3 core node backbone is that each node has
> but two paths. And I believe that each end node has but one path into and
> out of the MCI national backbone. Now I am new to networking so there may
> be something that I am missing here, but as I remember every node except
> one on the old t-1 backbone had THREE paths to the rest of the network,
> the one that did not had two paths (the midnet node I believe).
>
> Would it be correct to conclude that such a network topology is MUCH more
> failure prone than the T-1 net that is about to be turned off.
>
> How DOES the community benefit from the new architecture and the increased
> speed?? Can someone show me what I am missing?
Gordon, the primary design criterion (and this was also prior to ANS (BA?))
was increased robustness. Also it had to be affordable. Multiple tail circuits
to individual sites are extremely expensive (hurt with T1, kill with T3!).
Those multiple tail sites would likely go through the same conduit or uWave
link, and a cut would have likely broken all tails. We had also experienced
some power and operator support problems at some sites. So, a principal
consideration was the survival of the network without serious performance
or connectivity degradation. It appeared sense to co-locate the nodes in
places where there was sufficient UPS power, constant operational support,
and at centerpieces of networking bandwidth. So, MCI junction points were
the obvious choice. In 1987 MCI was not read for co-location, but in 1989
and 1990 they were quite receptive to the idea. This is also where the
cloud model came from, namely that the MCI junction points have robust
connectivity to their national (two-dimensional, rather than individual links)
high speed network (circuits). They connectivity could even be switched between
junction points in case of catastrophic events. So, the connectivity between
the core sites, while given at any point of time, was considered as changable
at any point of time, not unlike the earlier plans for Digital Reconfiguration
Services for the MCI circuits on the T1 network (a plan that had been
abandoned when the focus on the T3 network started). In any case a robust
"cloud" network was designed and implemented. SIngle tails feed the sites,
not the least because of redundancy on tail circuits being a bit of an illusion
anyway. The close collaboration with phone companies these days, which wasn't
quite there in pre-NSFNET days, allowed for such considerations of some higher
integration of services. My hope is that phone companies will do much more
of that, actually, as they are in *the* strategic position to provide needed
services in the longer run!
>From: "Erik E. Fair" (Your Friendly Postmaster) <fair@apple.com>
>
>One other thing about the distinction between the T1 and T3 NSFNET
>backbones, to emphasize Rick Adams' point - I find it significant that
>the T1 backbone exists wholly within the premeses of the various
>regional networks, redundancy and all, whereas the T3 network is built
>from Core NSS's, located in MCI POPs, and End NSS's, located at the
>regional network access point (e.g. Stanford for BARRNET). There is,
>according to my information, exactly one T3 link between any E-NSS and
>C-NSS, and no direct links between any two E-NSS's. This means that the
>T3 backbone, from the point of view of a regional network, has
>considerably less redundancy than the T1 backbone.
>
>The T1 backbone is clearly in the hands of the regionals, physically.
Erik, I don't quite believe this is true, for the reasons mentioned before.
For site reasons the T1 network (at least initially) was at times less robust.
Also note that the NSFNET node was never part/property of the site, but always
"owned, managed and operated" by the NSFNET partnership. Not that we could
have done without the great collaboration we got from all the sites with this,
in reality, joint endeavour, but still, the nodes did not "belong" to the sites.
>The T3 backbone is quite clearly owned and operated by ANS. Perhaps the
>regionals should consider keeping those "old" T1 links, and replacing
>the T1 NSS's with real routers when the NSF wants to dismantle the T1
>NSFNET backbone in favor of the T3 backbone. After all, in networking,
>redundancy is good.
If there is a robustness problem then it would be fixed. You cannot fix it
with a link matrix between the 4000+ networks in the system.
Some of the flames about the T1 routers are somewhat unfair. Yes, by todays
standards there are higher performance and more compact routers. But keep
in mind 1987. Things have really evolved since. I do not believe there was
a good alternative in 1987, which is why the multi-RT T1 routers got created,
to conform to performance consideration that NSF had expressed in the
1987 solicitation. The T1 NSS were great for that time.
Hans-Werner
