[429] in Commercialization & Privatization of the Internet
Why commercial sites aren't on the net
daemon@ATHENA.MIT.EDU (Morten Reistad)
Sun Mar 24 07:25:46 1991
Date: 24 Mar 91 12:05 +0100
From: Morten Reistad <MRR@boers.uu.no>
To: com-priv@uu.psi.com
Cc: mrr@boers.uu.no
I have been looking at the postings in this list for a while,
and have noted a roaring silence from all commercial interests.
This will break that silence.
First, to introduce myself. I am Morten Reistad, 30'ish, from
Norway (that is in Europe), and my background is 50/50
economics/finance and CS-related toptics.
I work for a very odd company; it is a limited company fully
owned by the Stock Exchange, with the responability to handle
external customer/network relations. It may formally be
Public Sector, but it sure does not look that way.
This is a really good place to observe the state of the commercial
networking scene, as almost everyone has a link to the Stock
Exchange. The "networking" is not a pretty picture. The OSI-
model is far, far away from the mindset of commercial implementors.
The networks are all X.25, SNA, Circuit-switched or timeslot-
multiplexed. Upper-layer protocol standardization is largely
unknown. Some of the LARGE operators must be pinched hard in the
arm to use anything else than asynch modems, and simplex, non-
error-correcting broadcasts. Some of them are exposed in the
USD 100M range on these wires.
We may therefore have an application to add to the "big three"
IP-users: Stock tickers. They consist of multicasting huge amounts
of variable updates. UDP and a registry service would do the
support job, and the protocol itself would be somewhere between
FTP and telnet in complexity.
However, first we must do some network building. The sorry state
of local networks was mentioned above.
Trips abroad lately have assured me that the jumble is even worse
elsewhere in Europe. (US comments would be welcome).
I see the internet suite of protocols as a good alternative
to these problems, but it will need a few amendments.
(Currently it is the ONLY implemented alternative for open
systems, as OSI systems are buggy, expensive, requires
dedicated wires and telco services, and no thought has
been given to implementing worldwide connectivity.)
The one good thing about IP here is that it can run on anything
from carrier pigeons to FDDI, and therefore is not dependent
on special hardware or services. This last point is often over-
looked in the commercial world, but the value of that freedom
for the corporate world is immense. It is the freedom to buy and
sell bandwitdh from whoever you choose.
The political side of the Internet is however not beneficial
to commercial interests for the moment.
This "careful use" policy means that the Internet cannot be
used for anything serious for us in the commercial sector,
except for some academic contacts on the r&d side, and for
some standardization work. We therefore operate what is in effect
an "Internet Island", with a uucp canoe for mail transport to
the mainland. Only "worthy" travellers can go in this canoe.
I find this policy getting to be demeaning and intimidating.
I want a net to PAY FOR and use as i d**m please. Nothing else
is good enough for a viable network. I get the same signals
from other friends and collegues in other companies. The best
net on the planet is about to be ruled by fear and arbitrary
rulers, because it is subsidized, and you need to show your
"worthiness" to receive that subsidy.
This is the reason we use lots of TCP/IP and have small canoes
to the real Internet.
Why is the net set up like it is ?
It is set up as a single, private net, with an abundance of
grey legal areas. It is a (natural and de-facto) monopoly,
but we are not regulating it. It is not a carrier, but it
carries lots of third-party traffic. Does it guarantee
privacy ? Is it immune from possible criminal traffic on
the net ? Since it is not a carrier, how ?
The legal and administrative shortcuts are about to bite us.
We have to do it the way the legal framework is set up.
That means separating the CARRIER(s), the central ADMINISTRATION,
REGULATORY AGENCIES must be set up, and DEMARCATION POINTS defined.
Otherwise we will let this great net vanquish as a public-sector
good reserved for the rich, strong and few. Try selling that to
your congressman.
The answer lies in breaking up the net in areas that each makes
commercial sense, and can pay their way in the world. Note that
this can be done with minimal impact on the existing network.
To do this we need to separate the "netcore", i.e. the central
nameservice from the operators and the carriers. The carrier
services must also be explicitly removed from the network.
Today the net is a mess of all three.
Since the net is basically constructed right, carrier services
can be set up easily. This means just defining which layer you
want to set the demarc at, and then for the carrier to comply
with the legal requirements to be a common carrier. This demarc
may differ from carrier to carrier.
The core of the service is a natural monopoly, (There can only
BE one set of root nameservers, right ?) This must be handled
as we handle all other such monopolies, preferably by some federal
or international agency, or by a private bidder on a contract from
such agency.
I would see it as natural for them to price their services at a level
where the revenues would cover cost, plus possibly some profit to
make it interesting to bid for the contract.
None of these would have any impact on the design of the present net,
but it would open the door wide open to a great commercial network
as well, basically existing in paralell with the present net,
touching where this would be natural.
Then, onto the Internet amendments.
Carriers are notably absent in the present Internet, even
though the present network design almost requires them.
This has led to a grand demand for class B and C networks,
but there is a huge underutilization of the class A network
space. (How many class A hosts are there, actually ?)
Since the IP protocol was made to run on top of something else,
ethernet, arpanet, or whatever, the absence of underlying
long distance protocol has become a problem. We currently use
IP instead of having functionality for local switching in the
underlying layer. I.E. We need a functional replacement
for the Arpanet. There are two contenders as I see it,
X.25 and frame relay. The first one has severe performance problems
and resides too firmly in the telco world, and the second one
is not quite here yet. This is however an issue that has received
far too little attention. We may have to form a "commercial
internet carrier task force" to set up subnets for our ip-packets.
I suggest that we start activites to use the existing X.25
networks, and set up a class A network on top of the global
X.25 network, with a decent address resolution service to navigate
this network. This can be operated on the side of the Internet
Proper, as long as the router manufacturers will support the
address resolution used. Didn't I see some A-net for X.25 somewhere ?
Note that this is an activity that can be performed completely on
the side of the existing Internet, except for some core entries.
We may have to do a similar thing for Frame Relay. Other local
networks like DATEX (fast, cheap 9600-baud circuit switching) may also
qualify as low-level protocols worthy of an ARP-service. ISDN may
also be ready for this soon. None of these networks will take
off as carriers for IP before we have a protocol to navigate them.
All of this is needed to bootstrap a commercial net using
existing networks as carriers.
I do, however, still sense a void in the just-below-ip strata in
the protocol stack. Isn't there a need for a simple connection-
less packet-mover that can accomodate moving IP packets on a pure
link level, arp-driven low-to-medium speed net, based on standard
leased/switched circuits ? The popularity of non-protocols like
SLIP should attest to this. Just something to carry IP frames
without costing a bundle.
A redesign of the ARPAnet after 25 years of added experience
may not be such a bad thing, to function as a network platform
with 3 to 100+ hosts constituting a subnet the rest of us can use.
The next issue is that this needs to be a truly international
network, not just a jumble of national patchwork. This may be
more tricky to acheive. Ideas anyone ?
After all of this we may just have constructed a network that anyone
in the reasonably connected world may use and pay for.
Morten Reistad
System Designer
Oslo Stock Exchange Information <mrr@boers.uu.no> +47 2 341 700
