[596] in Commercialization & Privatization of the Internet
Re: At What Price Will TCP/IP Connections Gain Wide Market Appeal?
daemon@ATHENA.MIT.EDU (John Gilmore)
Tue Apr 16 07:25:28 1991
Date: Mon, 15 Apr 91 23:50:06 PDT
From: gnu@toad.com (John Gilmore)
To: com-priv@psi.com
At a few hundred dollars a month, lots of small businesses will switch
to TCP/IP. They're already paying that for uucp, express package
delivery, fax, etc, which TCP can do if their customers are willing to
hook to it. (A number of our customers are getting on the Alternet; we
had to steer them away from the BARRNET so they could pass commercial
traffic with us, but it turns out Alternet rates are lower for
businesses anyway, since Alternet is not trying to rob Commercial Peter
to pay Academic Paul, but charges everyone equal rates for equal
service.)
It is already possible to pay monthly prices on the $2-400 order, for 56kbit
service. The catch is that you have to pay a lot up front for the low volume
hardware that handles 56k service. This includes Telco installation ($2000),
"DSU" modems ($600 x 2), and a router at both ends ($3000 x2 -- and up).
Once you have laid out $9000 in capital costs, sure, it's only a few hundred
bucks a month...
The charges for connecting to a network access provider such as
Alternet or PSInet can be split down N ways by leasing your own lines
among friends, in the grand uucp tradition. Alternet has no third
party traffic prohibition (Thanks Rick!) and for the other companies
the extra cost is less than everybody hooking up separately. You get
less service (it goes down more, you share the bandwith, you
administer, etc), but it's cheap.
We can reduce the startup costs in various ways. Cheaper data service
via ISDN tariffs would get rid of most of the $2000 installation, but
this is unavailable in many parts of the country&world, and there is
little equipment that can interface to it. The DSU's are already
cheaper than ISDN adapters (which tend to run $1200 and up), and are
competitive with 9600 baud dialup modems at $500 and up. The router is
where substantial money can be saved today; an IBM PC at $1000 including
Ethernet board and 56K serial board would work just fine as a router,
using public domain software such as Phil Karn's KA9Q. (Phil will want
$60 from you if you're commercial.)
At the moment it's one of those "for want of a nail" situations. We have
such a setup running here, connecting four sites. The IBM PC has a poor
serial chip that only runs async and at low speed, so you need an extra
board for sync serial. The boards are mostly poor or expensive; we found
a cheap one but don't have anyone who has the motivation, time, and expertise
to get a driver working in the incredibly cruddy DOS hardware/software
environment. PC DMA is so poor as to be unusable, so you have to field 7000
interrupts/second. Our current result is that we run 9600 baud SLIP
over our 56K modems until somebody shows up with a nail.
We could've used ISDN but for similar "nail" problems. Every Sun
SPARCstation comes with an ISDN chip in it (the sound chip), and early
SS-1 boards even had places where you could stuff $5 worth of passive
components and hook up ISDN on those four extra pins on the "audio"
connector, but one normally smart person in the Sun networking group
went braindead and refused to allow software to be written that would
use this hardware for networking. With no software support, the
hardware people took out the space for the passive components
connecting up this chip to ISDN lines, so it's only wired for audio
now. Eventually Sun will produce an expensive S-bus board for hooking
up to ISDN, but then we'll be back to $600 or more per end for
interfaces, equivalent to DSU's. Even if SPARCstation ISDN was
working, the local telco botched its installation and it only works in
a few places -- no demand, why bother with supply?
Someone could probably build a great router quite cheaply -- hell, the
SPARCstation SLC has all the hardware you need, 8MB of RAM, hot
Ethernet, ISDN, fast CPU, two serial ports capable of easy 56K sync
operation, and sells today for $3000 including a lot of fancy extras
like CRT, keyboard, SunOS license, etc. The SLC board itself is tiny,
like a PC add-in board; you could put it and a power supply in a box
the size of the average IBM PC shrink-scrapped software box, probably
for $1500. But it seems that nobody in the router market will get the
volumes that made the SLC worth designing; and nobody building high
volume products has put in the extra .07% effort to make their product
useful as a cheap router. (Sun won't even put SLIP or PPP into the OS,
though it's free; they make too much money selling proprietary
equivalents to people who don't know any better. Never mind that it
would expand their market further in the low end to have a $3000
workstation for small sales offices or home programmers that would
network at 128kbits/sec to the main office by plugging a modular jack
into a $12.95/month phone line.)
John Gilmore
