In the early years of the Net's development, the late '80s, the Internet
business outside campuses and corporations was a small-time and sometimes
tacky trade. In 1992, the entire Net comprised a million linked computers,
many of them in university and government labs. It wasn't until November
1993 that Net Access acquired a dedicated 56-kilobit line for direct connection
to an official network access point. Costing $400 per month, it multiplexed
22 dial-up modems among 250 users. With the Mosaic World Wide Web browser
yet to catch on outside the universities, Net Access did not even have
to supply SLIP (serial line interface protocol) or PPP (point-to-point
protocol) accounts, which shield the user from the details of Unix.
Freedman, however, saw the need for new technology to link people to
the full resources of the Net without having to know abstruse Unix commands.
"As a professional geek, writing code is my true calling," he
says, adding that he threw himself into this work. Although the program
was eclipsed by Mosaic, Lynx and other approaches, he still believes that
his software provided easier access to the Internet, complete with the
ability to trace routes and "ping" remote machines. Enabling
users to log in to the program in 1992, he put Net Access on the technological
forefront of ISPs.
The largest challenge for an ISP, then and now, is managing the floods
of bits engulfing a Usenet news server at a rate of some 500 megabytes
per day, five news articles per second, each with a unique identification
that has to be scanned to assure that the news is fresh and not duplicated.
The heart of the Internet until the arrival of the World Wide Web -- and
still cherished more than the Web by many Internet veterans -- Usenet is
the huge collection of textual bulletin boards and other information troves
and exchanges from which the communities of the Net exfoliate. As Steve
Willens of Livingston Enterprises puts it: "This is the real source
of the Internet as we know it and the challenge that forced the development
of technology specialized for the Net" -- notably Livingston communications
servers that linked modems to the Net through fast comports functioning
with compression at 115.2 kilobits per second.
In 1994, Freedman recognized he had a major business on his hands.
He decided to lease a T-1 line from PREP-NET (Pennsylvania Research and
Economic Partnership Network), which required a prepayment of $1,000 per
month. With 50 phone lines and modems and 500 users, he broke all ties
with Stony Brook and began hiring people to handle a rising tide of traffic
and a surging demand for technical support.
That summer, he had three full-time people: "Myself, my wife,
Gail, and my 20-year-old brother, Noam. Working with him made me realize
why people pay me so much money as a consultant [up to $150 an hour]. He
served as a kind of Avi echo, intuitively knowing what I wanted and when."
A student in computer science at the University of Chicago, Noam is in
the process of extending the business to that city, while Avi has established
points of presence in New York and Washington, D.C. He has hired five Net
Access customers, none with college degrees, to provide technical support
full time as the number of users has climbed at a pace of some 15% per
month since the end of 1994.
For links to other cities, Freedman relied on advice from telecommunications
consultant Gordon Jacobson, a Penn alumnus who maintains close links to
the Penn school of engineering, where his father graduated. With Jacobson's
help, Freedman is ending 1995 with a fiber circuit connecting him to MAE
East at 45 megabits a second, a 10-megabit-per-second link to Sprint's
network-access point, and more than half a dozen point-to-point T-1 lines,
all for well under half of the normally tariffed prices for these services.
With increasing broadband connectivity, Net Access commands more than half
as much bandwidth at the nerve centers of the Net as Netcom, which has
50 times more customers.
Though indispensable, technology alone cannot sustain a successful
ISP. It is people that make the vital difference. If Freedman had originally
hired people to perform the work that he did himself part-time -- "keeping
the machines running, maintaining software, recovering from disasters,
installing and tuning equipment and circuits" -- he would have incurred
expenses of some $100,000 per year and his financial model would have collapsed.
The reason many corporations are so slow to develop Internet programs is
not the lack of equipment but the dearth of personnel. The large companies
pursuing Net Access did not care about Freedman's rooms full of gear. They
were after Freedman himself.
Freedman's entrepreneurship and technology ride on a tide of other
enterprise by the suppliers of Internet gear. These, too, are not huge
telephone company equipment manufacturers or rising software monopolists
but mostly small or medium-size companies, led by young entrepreneurs,
fighting to survive in the most intensely competitive arena of the world
An Internet service provider must begin by supplying modems through
which the outside world can connect to his offerings. With millions of
home customers who dwarf the ISP modem volumes, U.S. Robotics is currently
ascendant in most ISPs, but Freedman spurns them for cheaper devices from
Multi-Tech. These modems connect to a Xylogics terminal server that authenticates
the name and password combination entered by the user and validates the
caller as legitimate. Then the customer enters Net Access's local-area
network linking a set of Sun Microsystems servers that supply World Wide
Web, Gopher, Usenet, e-mail, file transfer, Telnet and other Internet services.
Net Access is unusual for an ISP, since few use Xylogics equipment.
Recently bought out by Bay Networks, Xylogics supplied nearly all the terminal
servers for the university market, and it still shies away from the tumultuous
world of ISPs. These customers mostly use Livingston products that run
a security protocol named Radius (remote authentication dial-in user services).
Channeling the bits around the ISP's internal net and on to other networks
are banks of routers, also often built by Cisco or Livingston (although
Freedman originally chose Morningstar because it was cheaper). Linking
a particular ISP to other ISPs and network access points are T-1 cables
running at 1.544 megabits per second through multiplexing and demultiplexing
and conditioning equipment. These functions are performed by DSU-CSUs (data
service units-channel service units) made by such companies as TxPort,
Adtran, General DataComm and ADC Kentrox.
Freedman insists on the Law of the Microcosm in choosing all his equipment
and in making all his projects for expansion. Since his study of distributed
computing at Temple, he has everywhere cherished duplication and redundancy
and cheap components over centralization and scale economies. He at first
bought a nine-gigabyte drive from Micropolis. Now he regrets the decision
and is replacing it with five two-gigabyte drives (more I/O [input/output],
redundancy and reliability). "The more spindles the better,"
he says. He buys lots of cheap secondhand Suns rather than one powerful
server. He criticizes some of the larger ISPs, such as Netcom, for centralizing
their servers and technical support. It causes bottlenecks and delays,
he says, and opens the system to crashes if any of the communication lines
Freedman's rule is to provide service as locally as possible. He believes
ISPs with fully equipped local network sites, rather than mere communications
nodes like Netcom's, will prevail. Like most small ISPs, Freedman is wedded
to flat-rate pricing, though his accounts of altercations with customers
who want to resell or overgraze his commons may undermine confidence that
this pricing regime can survive into the future. But managing flat-rate
prices is a core competence of the ISPs. Believing that bits will flee
toward flat rates, Freedman says MCI will fail in its plans to transform
Internet pricing models by adding some as yet unannounced scale of measured
usage based on time, packets or both.
Is Freedman's model scalable, or is it doomed as he grows? Could Freedman
be displaced by MCI or Sprint-Comcast or Bell Atlantic or Microsoft-UUNet
or AT&T in a siege of merger-monopolization? He believes that up until
a threshold of some 25,000 to 50,000 customers, meaning revenues of between
$5 million and $10 million net of more lucrative business clients, his
economic and technical model can trump all comers. At that point, he will
face the usual entrepreneurial crisis of transition: Freedman will need
business partners, routinized technology management schemes and expensive
accounting to maintain operations as Net Access spreads across the country.
But he does not fear competition. His problems, he says, are servicing
the flood of new customers and anticipating the depredations of "Congresscritters"
who want to make him liable for any vagrant flasher who strays onto one
of his hard drives.
Still a small force in the global matrix of telecommunications, Freedman
now dreams of exploiting available resources of fiber, dark and lit, to
acquire major new bandwidth, linking cities up and down the East Coast
and across the U.S. Helping Freedman move this project toward reality is
his telecom guru Jacobson, an entrepreneurial dervish from Portman Communications.
With financiers on the line to supply some $5 million in startup capital,
Jacobson is planning to launch a national IRamp network. The service will
ultimately open fully staffed Internet access facilities in 30 cities nationwide,
linked everywhere by fiber, at a cost of some $1 million per site.
Such investment looms large compared to the rock-bottom base of Freedman's
operation, and easily eclipses a national ISP's point-of-presence facility
that can cost upwards of $70,000. But David Farber, gigabit-testbed guru,
recently told a New York audience at the Penn Club that, spurred by business
needs, the marketplace is seeking higher-end, stable-broadband ISP services
that can handle millions of hits a day at a Web site with no access delays
or congestion and that provide local access and custom software configuration.
For these high-end customers, the SPARC 20 servers and T-1 and 56-kilobit
links of the many small ISPs will no longer suffice.
Pioneering the kind of broadband channels that will eventually become
ubiquitous on the Net, IRamp's planned facilities will command OC-3 fiber
(155 megabits per second) links to a national network of both dark and
lit fiber, available from utilities, pipelines and other unusual sources.
Such bypass strategies will become increasingly common in coming years.
The 10 million miles of fiber currently installed in the U.S., after all,
is exploited to approximately one-millionth of its potential capacity --
and much of it is unused "dark fiber."
For key ISP server and security functions, Jacobson plans to use fully
fault-tolerant Tandem S4000 servers running the new ServerNet multibus
scheme. It was conceived by venerable Tandem designer Robert Horst as a
new-generation architecture explicitly optimized to substitute bandwidth
for switching speeds. Fully scalable, ServerNet was licensed in October
by Compaq, yet it commands a theoretical throughput limit of an unprecedented
petabit per second (a million billion bits). For graphics-intensive applications,
Jacobson envisages Silicon Graphics WebForce Challenge S servers using
Irix software. Even with as few as 5,000 subscribers per site paying a
competitive nonusage-based rate, Jacobson projects a high rate of return.
Meanwhile, at Netcom, the nation's largest ISP, David Garrison, the
CEO, is undergoing the stresses that Freedman foresees for himself as he
expands his business. During his previous stint at the helm of the meteoric
paging company, SkyTel, Garrison, a rangy dark-haired entrepreneur with
a slight uneasiness in his ready smile of prosperity, thought he had approached
the ultimate in entrepreneurial excitement. But nothing in his career in
the wireless industry prepared him for his first nine months as head of
Netcom. Here is a company that during the last three quarters grew from
400 to 1,200 employees, from 58 to 201 points of presence, from 72,000
to more than 200,000 customers, and from revenues of $12.4 million in 1994
to a $50 million run rate in 1995 and to a market cap of some $400 million,
while the traffic in bits grows at an even faster pace -- impelled by the
graphic demands of the World Wide Web, itself expanding at the rate of
more than 1,000 new servers per week.
Netcom pares down its points of presence to simple communications nodes
and handles all the technical support and Internet services for them at
the company's headquarters. This operation fills up a high-rise in San
Jose. Some floors teem with desks manned by earnest engineers in jeans,
many of them Asian, working the phones. Other floors are replete with row
upon row of racks filled wall-to-wall with Cisco routers, Sun servers,
Livingston PortMasters, Ascend ISDN pipelines, Cascade edge switches and
U.S. Robotics modems. Walking through these ever-expanding mazes of machinery,
Garrison's entrepreneurial smile at times moves from the ready to the giddy.
In this environment of riotous growth, the telcos move their slow thighs
like trolls under the bridges and routers of the Internet. Currently commanding
perhaps 2% of the traffic, AT&T, for example, has declared its ambition
to capture 60% of the Internet business over the next two years. But Garrison
demurs: "From the Olympian perspective of a McKinsey & Co. consultant,
AT&T could take over any business. They have one of the greatest brand
names in the world, they've got more money than God, a billing relationship
with some 40 million people, a global network and alliances and consortia,
Internet pioneer Bolt, Beranek & Newman in their fold, and they have
perhaps the world's largest internal World Wide Web on their own Unix servers
among their 300,000 employees."
But like most of the telcos, AT&T lacks focus. As Netcom marketing
chief John Zeisler explains: "Phone companies have their 700 numbers,
800 numbers, corporate customers, their Hollywood links, their leased lines,
their frame relay, their ADSL (asymmetric digital subscriber line), their
cable aspirations, their huge wireless opportunities, their bureaucracy,
their regulatory tariffs, their pricing confusions. Should voice be priced
as data or should data be priced as voice? They are great at laying fiber
and wire, connecting it to switches and bringing signals to the central
office and to the curb. But the Internet is a second thought, just another
business to them."
As in the PC industry, focus and agility are crucial. In an arena where
the technologies ride a remorseless onrush of exponential changes, no prolonged
bureaucratic process can succeed. Even the maps and schematics of rapid
convergence among media industries miss the point. Dominating this arena
is the computer industry -- with its millions of piranha processors and
entrepreneurs -- and it doesn't converge with anything; it eats everything
in its path.
Now ascendant is the Internet computer industry. Most of these new
companies, from Livingston to Netscape, focus on the Internet. Using personal
computer components to reduce the price of ISP infrastructure far below
the price of telco installations, these companies endow the ISPs with a
further advantage in a dynamic industry.
Livingston Enterprises epitomizes the success of the new companies
creating this new industry. Secreted in Pleasanton, Calif., and financed
by corporate cash flow, Livingston has grown up with the Internet at a
pace not far in the wake of its more illustrious rival, Cisco Systems.
Livingston PortMasters crowd Netcom's headquarters, as they do most of
the other ISPs.