20 Years of IT History: Connecting Devices, Data and People
The story of the past 20 years of technology has been all about connecting the dots between computers, data and the people who use them.
By Fred Hapgood
Throughout the late 1980s, microcomputers had been
understood as toy versions of mainframes and minis; they were
standalone devices that attacked problems with processing
cycles. In the quaint locution of the day, they were
Gradually, the devices acquired a different function. They
became smart links, machines that connected devices, data and
people. They went from being computing machines to connection
machines. Much of the history of the last 20 years can be
written in terms of who got this and who did not.
IBM’s rollout of its PS/2 microcomputer came on two
levels, both news. The ads raved about the classy technical
specs: a blazingly fast internal architecture, plug-and-play
BIOS, keyboard and mouse interfaces that are still in use today
(and are still called the PS/2 interface) and a floppy disk
format (1.44M) that was so good it lasted as long as the
The analysts saw a different message: IBM had decided to
shoo the children away. Where the PC had been wide open, the
PS/2 was buttoned tight. Every aspect of it was proprietary,
including the operating system. Businesswise, the job of the
PS/2 was to yank the rug out from under both the clone
manufacturers and that upstart, Microsoft.
There was no reason to bet against IBM. It had the classiest
brand, an immense promotional budget and some of the best
engineers in the world. Yet, incredibly, after several years of
very expensive triage, the PS/2 initiative crashed and burned.
The failure was a body blow to IBM and its standing in the
What went wrong? The fingers of blame pointed in every
direction (silly ads, pricing), but the truth is the PS/2 was
the wrong product for a market coalescing around connectivity.
Sizzling performance is nice but not essential in a connector
because performance is measured against the entire system, not
any one part. Blatant assertions of ownership—this is my
toy—threatened compatibility, the key virtue in a
IBM failed to understand the important difference between a
connection machine and a computing machine. And it paid the
1988: Next and OOPs
The connectivity story continued with Steve Jobs’s Next.
When you bought a Next, you got a piece of great (but
completely closed) hardware (which, of course, looked totally
cool) and an operating system built around a programming
philosophy new to micros: Object Oriented Programming.
Where traditional programming focused on logical operations
(computing), OOP’s great strength was the management of
categories or classes, including hierarchies of classes. This
made it possible to write programs that pulled more kinds of
things (humans, structures, classes, data) into a given
computing environment without forcing the programmer to rewrite
these environments from scratch. OOP was a programming language
for connection machines.
The totally cool but closed hardware totally failed, while
OOP went on to become bigger than the Next computer itself
could have ever been. Today, many important computing languages
(Java, C++, PERL, SmallTalk, among others) come with an OOP
1989: Netware 3
The first customers of micros, largely programmer types, found
ways to use their new toys on the job. As collections of these
machines aggregated at various institutes and centers, the idea
inevitably occurred to their owners that it would be neat to be
able to hook everybody’s micro together (and to the main
With every passing year, the amazing power of connectivity
was becoming more evident. At least in theory, every device you
plugged into the network inherited the assets and resources of
every other machine on that net. In 1980 the inventor of
Ethernet, Bob Metcalfe, took a stab at quantifying the gains to
networking by proposing a law that the utility of a network
went up with the square of the devices connected to it. While
people did and do argue over whether Metcalfe got the exponent
precisely right, nobody doubts that he nailed the spirit of the
But getting stability, predictability and compatibility out
of a grab bag of machines, themselves in constant flux, was not
easy. Novell had been working on the problem since 1983. By
1989 enough hair had been trimmed from the software that people
of reasonable skill could use it. Netware 3 was networking for
the rest of us, plus it was optimized for Intel’s very
popular 386 processor. As this combo spread throughout the
world it took with it the gospel of connectivity, leaving hosts
of beleaguered CIOs struggling to migrate their systems from
the quiet world of host/terminal to the mosh pit of
In the early years of the Internet, the connection between
users and resources was quite informal. If A wanted a specific
kind of file or program, he asked around, hoping that someone
had seen something like that somewhere and remembered the
address. If B wrote a cool program that she thought others
might like, she tried to find ways to spread the news This was
not ideal, but in the early days of the Net everybody knew
everybody else (practically), so the problem was not acute.
But by 1990 the community was expanding rapidly and finding
stuff was getting harder. That year three McGill students, Alan
Emtage, Bill Heelan and Peter J. Deutsch, attacked the problem
with a program they called Archie (from “archive”).
Archie worked by sending a message from your local system to
each entry on a list of servers, asking for the public files
available on that server. It would then combine the responses
into a single master list on your local system, which you would
then interrogate with “Find” commands. Archie was
crude, but it illustrated two big points about networking.
First, connectivity is self-extending; it creates entirely
new objects, which can themselves become subject to
connectivity. And if you connect A, B and C, you can create AB,
BC, AC, ABC and so on. These newly created objects might be
more useful than A or B or C. The master list generated by
Archie was the first step in the evolution of the Internet from
a network of networks to a library of resources.
Second, on a network, digital resources can be reused, over
and over, forever, at next to no additional cost. Put a search
engine on that network and you allow this efficiency to scale
without limit. This fact would turn out to have huge economic
A student at the University of Helsinki named Linus Torvalds
released a half-finished operating system, hoping that a few
hands might be willing to help out. To his surprise, he found
hundreds and then thousands of programmers willing and able to
work on the program, which he named Linux. As it turned out, a
large network is perfect for supporting projects that are
themselves networks, projects made up of pieces that can be
worked on in isolation and then combined…over the network.
These types of enterprises are enormously efficient, leveraging
small investments in time and energy by many people into highly
useful (and usually free) tools. Linux was one of the first of
these massively parallel collaborations, but soon enough they
would sprout up everywhere, from cartography
(“mashups”) to encyclopedias. And the Web
In 1992 Microsoft finally got a functional version of its
latest operating system out the door. Windows 3.1 advanced the
art in two ways; it was the first version to carry a useful
graphics interface, allowing inputs and outputs to be
represented and altered by manipulating icons. And more
important, Microsoft’s immense marketing power meant it
went on desktops everywhere in the world, becoming a de facto
Mosaic was released by the National Center for Supercomputing
Applications. What Windows 3.1 was to the microcomputer, Mosaic
was to the World Wide Web. Together, they acted to standardize
the Internet, allowing all the 3.1 installations (and other
compatible machines) to talk to each other with reasonable
levels of predictability and stability.
Metcalfe’s law is not automatic. As networks grow, the
potential to do more for less rises, but this benefit remains
theoretical until the network has passed through a phase of
greater formalization. As a systems scientist might put it, the
standardization of the core goes hand in hand with
differentiation of the edge. Each advances the other. No better
illustration of this point exists than the two episodes of
standardization above, which kicked off the immense flowering
of Internet content known as the World Wide Web.
As the Web appeared, seemingly out of nowhere, people became
convinced that something revolutionary was under way. CIOs
everywhere arrived at work to find a new item in their job
description: responsibility for getting their company a
website, beginning with registering the company’s name as
a URL, and weighing the delicate ethics of swiping those of
their more laggard competitors.
1994: Spam and More
Connectivity, we learned, has a dark side.
In 1994 two lawyers began advertising their services by
posting to Usenet groups en masse. They were widely reviled
(their ISP revoked their access), though in all fairness,
someone was going to walk through that door sooner or later.
“Green Card” spam (the villains were immigration
lawyers) was the opening gun of the age of malware for profit,
which eventually evolved into hundreds of flavors of spyware,
extortion schemes, Trojan horses, key loggers, zombies,
phishers, bots and so on. Today the average CIO probably spends
more time and energy worrying about blocking the bad that
networks can do than extending the good.
The Israeli company VocalTec announced Internet telephony and
RealAudio, streaming audio. These two announcements marked the
beginning of the great convergence carnival.
The VocalTec rollout presaged the struggle that VoIP was
about to catalyze between telecommunications and IT. The core
idea is that someday soon the network is going to eat it all
up—voice, music, video, news, data. Everything will be
connected to everything else. It’s inevitable, but that
doesn’t make dealing with the business, legal, political
and technological issues all this raises any easier.
1996: The Dotcoms
Sun Microsystems formed the JavaSoft group in order to develop
the Java technology. Java, a language optimized for writing
programs intended to run over a network, was (and is) a big
deal, but the news of the year was not technical but cultural.
This was the year when irrational exuberance slid behind the
wheel, the year the dotcom balloon broke free of its moorings
on planet Earth.
Much of the fever came from the spreading conviction that
old business models were dying: Why would anyone ever want to
go to a store anymore? How could a business compete if it was
carrying the overhead of a brick-and-mortar shop? All this
meant that anyone wanting a return on his investment had to
find a place to park it in cyberspace. Somewhere. Anywhere.
1997: Distributed Computing
Jeff Lawson of Distributed.net showed how the Internet could be
used to harness a very large number of geographically dispersed
microcomputers to attack a single problem—in this case, a
ciphertext released as a challenge by RSA (with a $10,000 prize
attached). Today distributed nets are being used to solve
problems in protein folding, the search for extraterrestrial
intelligence, financial modeling and many other problems. Under
the name grid computing, the concept has become a small but
important industry, offering companies needing lots of cycles a
cheap alternative to supercomputers.
XML, a markup language optimized for the Internet, supporting
most known human scripts and compatible across a wide range of
languages and platforms, increased the power and capacities of
1999: Wireless and Y2K
Time magazine named Jeff Bezos of Amazon its “Person of
the Year,” writing that “e-business is rapidly
replacing the traditional kind for almost any purchase you can
Also, on July 21, Steve Jobs demoed the first cheap wireless
modem. Wireless networking did not take the world by surprise.
For years everyone had understood that the need to embody
connectivity in physical wires was an immense constraint on the
growth of networking (and a fatal one, in the case of mobile
devices). People had been hammering away at the problem for at
least a decade, and a few very expensive solutions were running
here and there.
What was different about Apple’s AirPort was that it
was cheap enough for mass adoption. Over the next several
years, wireless LANs began to crop up everywhere. They
didn’t necessarily work perfectly; the technology came
with many headaches, beginning with security and dependability,
and CIOs were to spend many hours hammering out the bugs. They
did not, however, do much of that in 1999, for that year CIOs
were preparing for the imminent end of civilization, generally
known as the Y2K bug.
2000: Millennial Change and Angst
First, Y2K went off without a hitch, proving that luck is on
the side of those smart enough to be working on well-posed
problems and establishing ERP as the way businesses organized
Second, an Internet company (Google) developed a
well-grounded solution to the problem of making money over the
Third, the culture decided that the Internet did not mean
the end of business as we had known it, and everyone rose,
stretched and sold off their tech stocks. Good-bye, Pets.com,
Chipshot.com, and the rest.
People had been writing diaries on the Net for years, but the
form had never taken off. In the late 1990s, editors appeared
with several subtle enhancements, including browser-based
website editing, comments, permalinks, blogrolls and
trackbacks. These fixes turned Web diaries into blogs, and by
2001 blogs had become one of the great networking phenomena of
The development of blogs illustrates a subtle point about
connectivity. Conventional measurements of networks count nodes
and bandwidth, but connectivity has at least a third dimension:
adaptedness. Every object in a network has a trajectory of
enhancements that allow it to work better and do more in a
networked environment. One of the several ways in which
connectivity is self-extending is that it provides an
environment that selects for greater adaptivity to networking.
As objects move down this path, as they mature, connectivity
surges, even if nodes and bandwidth stay the same…which, of
course, they never do.
A number of accounting scandals from leading-edge tech
companies (Enron, WorldCom, and the like) led to legislation
designed to remake the financial reporting practices of public
companies from top to bottom. While Sox, as the act came to be
known, explicitly targeted the behaviors of CEOs and CFOs, it
probably changed the lives of CIOs as much or more.
Sox required that every act in a company’s financial
life be documented and that every document be auditable,
forcing CIOs to supervise a massive increase in documentation
and in the control of that documentation. Change management, in
particular, went from something the CIO could do on his or her
own in an afternoon (for reasons best known to the CIO) to an
agenda item for the Change Management Committee.
The scary part is that, given how integrated IT has become
with financial reporting, if a CEO or CFO were to be indicted
for Sox violations, the CIO is at risk of being sucked into the
same prosecution, as a coconspirator.
On the other hand, CIOs are now right at the heart of the
enterprise. The bean counters used to complain that IT was all
cost and no benefit. Thanks to Sox, IT can now point to a
benefit the most obtuse bean counter is likely to appreciate:
keeping him out of jail.
Imagine you have two (or more) IT objects, A and C. You want to
hook A and C together so they can send signals to each other.
Alas, they are incompatible, perhaps because they come from
different manufacturers. Virtualization is the business of
making a third object, B, that you slip between A and C to fool
each of the original objects into thinking the other is
speaking its own language, creating compatibility where before
there was none.
The IT objects could be anything at all: servers, operating
systems, routers, applications, hard disks, caches, whatever.
Virtualization allows you to hook anything up to anything else
and force the combo to work harmoniously.
Starting in 1999, a company named VMware committed itself to
the technology. The early years were slow. People complained
that everything had to be done twice (first by A or C and then
by B), which meant that everything took twice as many cycles
and burned up twice as many resources. The process added
complexity. But by 2003 the world was beginning to understand
how versatile and powerful a solution this was. One of signs of
this dawning comprehension came at the end of 2003 when EMC, a
huge storage company, bought a big piece of VMware. In 2007,
VMware went public and, in a generally listless market, had the
biggest tech stock debut since Google. Virtualization had
2004: ERP Hangover
In 2004 (or thereabouts), enterprise resource planning (ERP),
fell off the hype cliff and (perhaps this is the fairest way to
put it) became subject to the net of its positives and its
ERP is the art of framing a single formal definition for
every object and act in a company so that everything can be
managed together, top down. For instance, pre-ERP, each
department or division in a company usually defined the term
“employee” differently. These differences might be
tacit and hard to define and perhaps not even known to top
management, but they would usually matter. Once ERP came to
that company, “employee” would mean the same
everywhere, and every aspect of that identity would be explicit
and transparent. There would be one database for the entire
company and one interface to that database. A manager setting
policies for “employees” would know exactly what he
or she was doing. ERP is an instrument for bringing companies
to a higher degree of integration.
The great virtue of ERP lies in how well it supports
compliance with companywide policies. A given change just
radiates across the company, with every division learning about
it at the same time and in the same way. In the case of
Sarbanes-Oxley, which mandates a specific framework for
financial reporting, ERP seems essential to getting to
compliance at all.
All good. However, as experience with the technology
accumulated, downsides swam into view, among them a loss of
flexibility and weakest-link exposure risks—if one
department enters information inaccurately or imprecisely,
everybody suffers. There were others.
ERP is connectivity taken to the extreme; and while the
program has applications that are important and useful, it also
teaches that there are limits. Connectivity is not the solution
to all problems.
Sometimes it is even best avoided.
2005: Multicore Processors
AMD, the microprocessor manufacturer, announced the first
multicore microprocessor. Shortly thereafter, Intel follows
Multicore computing is understood as a new solution to the
problem of improving processor performance, but it might be
much more than that.
For decades computer scientists have known there’s an
alternative to traditional computing: having many processors
working on the same problem at the same time. But programming
for parallel processing is much harder than programming for a
single processor, and that difficulty has discouraged us from
exploring that technology.
Truth is, we never really had to go there because single
processors got faster so quickly, we never really needed an
However, ironically, it’s the need for processor speed
that is now forcing us to figure out parallel computing. The
faster processors run, the more power they consume and the more
heat they generate. Both of these are limiting factors. Because
multicore gets its edge by running more processors, not faster
ones, it allows the core to stay cool and energy-efficient.
Many analysts expect multicore to dominate processor design
from now on, with the number of connected cores per motherboard
rising steadily as we get better at solving the programming
problems presented by this new architecture. A decade from now
parallel programming will be the standard and perhaps we will
be a lot closer to matching the skills of the human brain.
2006: The Network
The growth in average traffic level (75%) outpaced the growth
of capacity (47%) on the world’s Internet backbones for
the third consecutive year.
For the last few decades, the world’s networking
engineers have done prodigies in building track just in front
of the advancing locomotive of Internet traffic. But recently
the locomotive has been gaining. We might be just a few years
away from a new kind of Internet, one in which applications are
triaged and bandwidth is metered and everybody has to make do
with performance levels far below ideal.
While there are many culprits, the biggest appetite out
there belongs to video. (In March 2007 more than 100 billion
videos were watched by users in the U.S. alone. That’s a
lot.) Almost every day brings news of a new Internet video
application, from movies on demand to TV-over-IP to civic and
or educational applications of YouTube. NBC has just announced
that it is planning to use the Internet to carry more than
2,000 hours of Olympics coverage next summer, and NBC will not
be the only network streaming from Beijing.
Comparable developments are unfolding in many enterprises as
video is used for more functions, from videoconferencing to
speeches by management to remote attendance at important
conferences. Eventually “bandwidth rationing” will
probably arrive even in companies with 10-gigabit LANs, and you
can guess on whose shoulders the task of imposing that
rationing is going to fall.
That’s right. The CIO.
2007: The iPhone
Steve Jobs demonstrates that he finally and totally and
conclusively gets the difference between machines that compute
and machines that connect.