Battle of the Networks
2022-08-07
Great read about the history of networks before "The World Wide Web".Chapter 2: Battle of the Networks from the book The Future of the Internet and How to Stop Itby Jonathan L. Zittrain
1) Book website
2) PDF from Harvard Libary
This work is licensed under a Creative Commons Attribution-Noncommercial-Share Alike 3.0 Unported License
Chapter 2: Battle of the Networks
As the price of computer processors and peripheral componentsdropped precipitously from the days of mainframes, it became easier
for computer technology to end up in people’s homes. But the crucial
element of the PC’s success is not that it has a cheap processor inside,
but that it is generative: it is open to reprogramming and thus repurposing by anyone. Its technical architecture, whether Windows, Mac,
or other, makes it easy for authors to write and owners to run new
code both large and small. As prices dropped, distributed ownership
of computers, rather than leasing within institutional environments,
became a practical reality, removing legal and business practice barriers to generative tinkering with the machines.
If the hobbyist PC had not established the value of tinkering so that
the PC could enter the mainstream in the late 1980s,1 what cheap
processors would small firms and mainstream consumers be using
today? One possibility is a set of information appliances. In such a
world, people would use smart typewriters for word processing from
companies like Brother: all-in-one units with integrated screens and
printers that could be used only to produce documents. For gaming,
they would use dedicated video game consoles—just as many do today. A personal checkbook might have had its own souped-up adding machine/calculator unit for balancing accounts—or it might have had no appliance at all, since
the cost of deploying specialized hardware for that purpose might have exceeded consumer demand.
There is still the question of networking. People would likely still want to exchange word processing and other documents with colleagues or friends. To
balance checkbooks conveniently would require communication with the
bank so that the user would not have to manually enter cleared checks and their
dates from a paper statement. Networking is not impossible in a world of
stand-alone appliances. Brother word processor users could exchange diskettes
with each other, and the bank could mail its customers cassettes, diskettes, or
CD-ROMs containing data usable only with the bank’s in-home appliance. Or
the home appliance could try to contact the bank’s computer from afar—an activity that would require the home and the bank to be networked somehow.
This configuration converges on the Hollerith model, where a central computer could be loaded with the right information automatically if it were in the
custody of the bank, or if the bank had a business relationship with a thirdparty manager. Then the question becomes how far away the various dumb terminals could be from the central computer. The considerable expense of building networks would suggest placing the machines in clusters, letting people
come to them. Electronic balancing of one’s checkbook would take place at a
computer installed in a bank lobby or strategically located cyber café, just as
automated teller machines (ATMs) are dispersed around cities today. People
could perform electronic document research over another kind of terminal
found at libraries and schools. Computers, then, are only one piece of a mosaic
that can be more or less generative. Another critical piece is the network, its
own generativity hinging on how much it costs to use, how its costs are measured, and the circumstances under which its users can connect to one another.
Just as information processing devices can be appliance, mainframe, PC, or
something in between, there are a variety of ways to design a network. The
choice of configuration involves many trade-offs. This chapter explains why
the Internet was not the only way to build a network—and that different network configurations lead not only to different levels of generativity, but also to
different levels of regulability and control. That we use the Internet today is not
solely a matter of some policy-maker’s choice, although certain regulatory interventions and government funding were necessary to its success. It is due to
an interplay of market forces and network externalities that are based on pre20 The Rise and Stall of the Generative Net
sumptions such as how trustworthy we can expect people to be. As those presumptions begin to change, so too will the shape of the network and the things
we connect to it.
BUILDING NETWORKS ON A NETWORK
Returning to a threshold question: if we wanted to allow people to use information technology at home and to be able to network in ways beyond sending
floppy diskettes through the mail, how can we connect homes to the wider
world? A natural answer would be to piggyback on the telephone network,
which was already set up to convey people’s voices from one house to another,
or between houses and institutions. Cyberlaw scholar Tim Wu and others have
pointed out how difficult it was at first to put the telephone network to any new
purpose,not for technical reasons, but for ones of legal control—and thus how
important early regulatory decisions forcing an opening of the network were to
the success of digital networking.2
In early twentieth-century America, AT&T controlled not only the telephone network, but also the devices attached to it. People rented their phones
from AT&T, and the company prohibited them from making any modifications to the phones. To be sure, there were no AT&T phone police to see what
customers were doing, but AT&T could and did go after the sellers of accessories like the Hush-A-Phone, which was invented in 1921 as a way to have a
conversation without others nearby overhearing it.3 It was a huge plastic funnel
enveloping the user’s mouth on one end and strapped to the microphone of
the handset on the other, muffling the conversation. Over 125,000 units were
sold.
As the monopoly utility telephone provider, AT&T faced specialized regulation from the U.S. Federal Communications Commission (FCC). In 1955, the
FCC held that AT&T could block the sale of the funnels as “unauthorized foreign attachments,” and terminate phone service to those who purchased them,
but the agency’s decision was reversed by an appellate court. The court drolly
noted, “[AT&T does] not challenge the subscriber’s right to seek privacy. They
say only that he should achieve it by cupping his hand between the transmitter
and his mouth and speaking in a low voice into this makeshift muffler.”4
Cupping a hand and placing a plastic funnel on the phone seemed the same
to the court. It found that at least in cases that were not “publicly detrimental”—in other words, where the phone system was not itself harmed—AT&T
had to allow customers to make physical additions to their handsets, and manBattle of the Networks 21
ufacturers to produce and distribute those additions. AT&T could have invented the Hush-A-Phone funnel itself. It did not; it took outsiders to begin
changing the system, even in small ways.
Hush-A-Phone was followed by more sweeping outside innovations. During
the 1940s, inventor Tom Carter sold and installed two-way radios for companies with workers out in the field. As his business caught on, he realized
how much more helpful it would be to be able to hook up a base station’s radio to a telephone so that faraway executives could be patched in to the front
lines. He invented the Carterfone to do just that in 1959 and sold over 3,500
units. AT&T told its customers that they were not allowed to use Carterfones,
because these devices hooked up to the network itself, unlike the Hush-APhone, which connected only to the telephone handset. Carter petitioned
against the rule and won.5 Mindful of the ideals behind the Hush-A-Phone
decision, the FCC agreed that so long as the network was not harmed, AT&T
could not block new devices, even ones that directly hooked up to the phone
network.
These decisions paved the way for advances invented and distributed by
third parties, advances that were the exceptions to the comparative innovation
desert of the telephone system. Outsiders introduced devices such as the answering machine, the fax machine, and the cordless phone that were rapidly
adopted.6 The most important advance, however, was the dial-up modem, a
crucial piece of hardware bridging consumer information processors and the
world of computer networks, whether proprietary or the Internet.
With the advent of the modem, people could acquire plain terminals or PCs
and connect them to central servers over a telephone line. Users could dial up
whichever service they wanted: a call to the bank’s network for banking, followed by a call to a more generic “information service” for interactive weather
and news.
The development of this capability illustrates the relationships among the
standard layers that can be said to exist in a network: at the bottom are the
physical wires, with services above, and then applications, and finally content
and social interaction. If AT&T had prevailed in the Carterfone proceeding, it
would have been able to insist that its customers use the phone network only
for traditional point-to-point telephone calls. The phone network would have
been repurposed for data solely at AT&T’s discretion and pace. Because AT&T
lost, others’ experiments in data transmission could move forward. The physical layer had become generative, and this generativity meant that additional
types of activity in higher layers were made possible. While AT&T continued
22 The Rise and Stall of the Generative Net
collecting rents from the phone network’s use whether for voice or modem
calls, both amateurs working for fun and entrepreneurs seeking new business
opportunities got into the online services business.
THE PROPRIETARY NETWORK MODEL
The first online services built on top of AT&T’s phone network were natural
extensions of the 1960s IBM-model minicomputer usage within businesses:
one centrally managed machine to which employees’ dumb terminals connected. Networks like CompuServe, The Source, America Online, Prodigy,
GEnie, and MCI Mail gave their subscribers access to content and services deployed solely by the network providers themselves.7
In 1983, a home computer user with a telephone line and a CompuServe
subscription could pursue a variety of pastimes8—reading an Associated Press
news feed, chatting in typed sentences with other CompuServe subscribers
through a “CB radio simulator,” sending private e-mail to fellow subscribers,
messaging on bulletin boards, and playing rudimentary multiplayer games.9
But if a subscriber or an outside company wanted to develop a new service that
might appeal to CompuServe subscribers, it could not automatically do so.
Even if it knew how to program on CompuServe’s mainframes, an aspiring
provider needed CompuServe’s approval. CompuServe entered into development agreements with outside content providers10 like the Associated Press
and, in some cases, with outside programmers,11 but between 1984 and 1994,
as the service grew from one hundred thousand subscribers to almost two million, its core functionalities remained largely unchanged.12
Innovation within services like CompuServe took place at the center of the
network rather than at its fringes. PCs were to be only the delivery vehicles for
data sent to customers, and users were not themselves expected to program or
to be able to receive services from anyone other than their central service
provider. CompuServe depended on the phone network’s physical layer generativity to get the last mile to a subscriber’s house, but CompuServe as a service
was not open to third-party tinkering.
Why would CompuServe hold to the same line that AT&T tried to draw?
After all, the economic model for almost every service was the connect charge:
a per-minute fee for access rather than advertising or transactional revenue.13
With mere connect time as the goal, one might think activity-garnering usercontributed software running on the service would be welcome, just as usercontributed content in the CB simulator or on a message board produced revBattle of the Networks 23
enue if it drew other users in. Why would the proprietary services not harness
the potential generativity of their offerings by making their own servers more
open to third-party coding? Some networks’ mainframes permitted an area in
which subscribers could write and execute their own software,14 but in each
case restrictions were quickly put in place to prevent other users from running
that software online. The “programming areas” became relics, and the Hollerith model prevailed.
Perhaps the companies surmised that little value could come to them from
user and third-party tinkering if there were no formal relationship between
those outside programmers and the information service’s in-house developers.
Perhaps they thought it too risky: a single mainframe or set of mainframes running a variety of applications could not risk being compromised by poorly
coded or downright rogue applications.
Perhaps they simply could not grasp the potential to produce new works that
could be found among an important subset of their subscribers—all were instead thought of solely as consumers. Or they may have thought that all the
important applications for online consumer services had already been invented—news, weather, bulletin boards, chat, e-mail, and the rudiments of
shopping.
In the early 1990s the future seemed to be converging on a handful of corporate-run networks that did not interconnect. There was competition of a sort
that recalls AT&T’s early competitors: firms with their own separate wires going to homes and businesses. Some people maintained an e-mail address on
each major online service simply so that they could interact with friends and
business contacts regardless of the service the others selected. Each information
service put together a proprietary blend of offerings, mediated by software produced by the service. Each service had the power to decide who could subscribe, under what terms, and what content would be allowed or disallowed,
either generally (should there be a forum about gay rights?) or specifically
(should this particular message about gay rights be deleted?). For example,
Prodigy sought a reputation as a family-friendly service and was more aggressive about deleting sensitive user-contributed content; CompuServe was more
of a free-for-all.15
But none seemed prepared to budge from the business models built around
their mainframes, and, as explained in detail in Chapter Four, works by scholars such as Mary Benner and Michael Tushman shed some light on why. Mature firms can acquire “stabilizing organizational routines”: “internal biases for
certainty and predictable results [which] favor exploitative innovation at the
24 The Rise and Stall of the Generative Net
expense of exploratory innovation.”16 And so far as the proprietary services
could tell, they had only one competitor other than each other: generative PCs
that used their modems to call other PCs instead of the centralized services. Exactly how proprietary networks would have evolved if left only to that competition will never be known, for CompuServe and its proprietary counterparts
were soon overwhelmed by the Internet and the powerful PC browsers used to
access it.17 But it is useful to recall how those PC-to-PC networks worked, and
who built them.
A GRASSROOTS NETWORK OF PCs
Even before PC owners had an opportunity to connect to the Internet, they
had an alternative to paying for appliancized proprietary networks. Several
people wrote BBS (“bulletin board system”) software that could turn any PC
into its own information service.18 Lacking ready arrangements with institutional content providers like the Associated Press, computers running BBS
software largely depended on their callers to provide information as well as to
consume it. Vibrant message boards, some with thousands of regular participants, sprang up. But they were limited by the physical properties and business
model of the phone system that carried their data. Even though the Carterfone
decision permitted the use of modems to connect users’ computers, a PC hosting a BBS was limited to one incoming call at a time unless its owner wanted to
pay for more phone lines and some arcane multiplexing equipment.19 With
many interested users having to share one incoming line to a BBS, it was the
opposite of the proprietary connect time model: users were asked to spend as
little time connected as possible.
PC generativity provided a way to ameliorate some of these limitations.
A PC owner named Tom Jennings wrote FIDOnet in the spring of 1984.20
FIDOnet was BBS software that could be installed on many PCs. Each FIDOnet BBS could call another in the FIDO network and they would exchange
their respective message stores. That way, users could post messages to a single
PC’s BBS and find it copied automatically, relay-style, to hundreds of other
BBSs around the world, with replies slowly working their way around to all the
FIDOnet BBSs. In the fall of 1984 FIDOnet claimed 160 associated PCs; by
the early 1990s it boasted 32,000, and many other programmers had made
contributions to improve Jennings’s work.21
Of course, FIDOnet was the ultimate kludge, simultaneously a testament to
the distributed ingenuity of those who tinker with generative technologies and
Battle of the Networks 25
a crude workaround that was bound to collapse under its own weight. Jennings
found that his network did not scale well, especially since it was built on top of
a physical network whose primary use was to allow two people, not many computers, to talk to each other. As the FIDOnet community grew bigger, it was no
longer a community—at least not a set of people who each knew one another.
Some new FIDOnet installations had the wrong dial-in numbers for their
peers, which meant that computers were calling people instead of other computers, redialing every time a computer did not answer.
“To impress on you the seriousness of wrong numbers in the node list,” Jennings wrote, “imagine you are a poor old lady, who every single night is getting
phone calls EVERY TWO MINUTES AT 4:00AM, no one says anything,
then hangs up. This actually happened; I would sit up and watch when there
was mail that didn’t go out for a week or two, and I’d pick up the phone after dialing, and was left in the embarrasing [sic] position of having to explain bulletin boards to an extremely tired, extremely annoyed person.”22
In some ways, this was the fear AT&T had expressed to the FCC during the
Carterfone controversy. When AT&T was no longer allowed to perform quality
control on the devices hooking up to the network, problems could arise and
AT&T would reasonably disclaim responsibility. Jennings and others worked
to fix software problems as they arose with new releases, but as FIDOnet authors wrestled with the consequences of their catastrophic success, it was clear
that the proprietary services were better suited for mainstream consumers.
They were more reliable, better advertised, and easier to use. But FIDOnet
demonstrates that amateur innovation—cobbling together bits and pieces
from volunteers—can produce a surprisingly functional and effective result—
one that has been rediscovered today in some severely bandwidth-constrained
areas of the world.23
Those with Jennings’s urge to code soon had an alternative outlet, one that
even the proprietary networks did not foresee as a threat until far too late: the
Internet, which appeared to combine the reliability of the pay networks with
the ethos and flexibility of user-written FIDOnet.
ENTER THE INTERNET
Just as the general-purpose PC beat leased and appliancized counterparts that
could perform only their manufacturers’ applications and nothing else, the Internet first linked to and then functionally replaced a host of proprietary consumer network services.24
26 The Rise and Stall of the Generative Net
The Internet’s founding is pegged to a message sent on October 29, 1969. It
was transmitted from UCLA to Stanford by computers hooked up to prototype “Interface Message Processors” (IMPs).25 A variety of otherwise-incompatible computer systems existed at the time—just as they do now—and the
IMP was conceived as a way to connect them.26 (The UCLA programmers
typed “log” to begin logging in to the Stanford computer. The Stanford computer crashed after the second letter, making “Lo” the first Internet message.)
From its start, the Internet was oriented differently from the proprietary networks and their ethos of bundling and control. Its goals were in some ways
more modest. The point of building the network was not to offer a particular
set of information or services like news or weather to customers, for which the
network was necessary but incidental. Rather, it was to connect anyone on the
network to anyone else. It was up to the people connected to figure out why
they wanted to be in touch in the first place; the network would simply carry
data between the two points.
The Internet thus has more in common with FIDOnet than it does with
CompuServe, yet it has proven far more useful and flexible than any of the proprietary networks. Most of the Internet’s architects were academics, amateurs
like Tom Jennings in the sense that they undertook their work for the innate interest of it, but professionals in the sense that they could devote themselves full
time to its development. They secured crucial government research funding
and other support to lease some of the original raw telecommunications facilities that would form the backbone of the new network, helping to make the
protocols they developed on paper testable in a real-world environment. The
money supporting this was relatively meager—on the order of tens of millions
of dollars from 1970 to 1990, and far less than a single popular startup raised in
an initial public offering once the Internet had gone mainstream. (For example,
ten-month-old, money-losing Yahoo! raised $35 million at its 1996 initial public offering.27 On the first day it started trading, the offered chunk of the company hit over $100 million in value, for a total corporate valuation of more
than $1 billion.28)
The Internet’s design reflects the situation and outlook of the Internet’s
framers: they were primarily academic researchers and moonlighting corporate
engineers who commanded no vast resources to implement a global network.29
The early Internet was implemented at university computer science departments, U.S. government research units,30 and select telecommunications companies with an interest in cutting-edge network research.31 These users might
naturally work on advances in bandwidth management or tools for researchers
Battle of the Networks 27
to use for discussion with each other, including informal, non-work-related
discussions. Unlike, say, FedEx, whose wildly successful paper transport network depended initially on the singularly focused application of venture capital to design and create an efficient physical infrastructure for delivery, those individuals thinking about the Internet in the 1960s and ’70s planned a network
that would cobble together existing research and government networks and
then wring as much use as possible from them.32
The design of the Internet reflected not only the financial constraints of its
creators, but also their motives. They had little concern for controlling the network or its users’ behavior.33 The network’s design was publicly available and
freely shared from the earliest moments of its development. If designers disagreed over how a particular protocol should work, they would argue until one
had persuaded most of the interested parties. The motto among them was, “We
reject: kings, presidents, and voting. We believe in: rough consensus and running code.”34 Energy spent running the network was seen as a burden rather
than a boon. Keeping options open for later network use and growth was seen
as sensible, and abuse of the network by those joining it without an explicit approval process was of little worry since the people using it were the very people
designing it—engineers bound by their desire to see the network work.35
The Internet was so different in character and audience from the proprietary
networks that few even saw them as competing with one another. However, by
the early 1990s, the Internet had proven its use enough that some large firms
were eager to begin using it for data transfers for their enterprise applications. It
helped that the network was subsidized by the U.S. government, allowing flatrate pricing for its users. The National Science Foundation (NSF) managed the
Internet backbone and asked that it be used only for noncommercial purposes,
but by 1991 was eager to see it privatized.36 Internet designers devised an entirely new protocol so that the backbone no longer needed to be centrally managed by the NSF or a single private successor, paving the way for multiple private network providers to bid to take up chunks of the old backbone, with no
one vendor wholly controlling it.37
Consumer applications were originally nowhere to be found, but that
changed after the Internet began accepting commercial interconnections without network research pretexts in 1991. The public at large was soon able to sign
up, which opened development of Internet applications and destinations to a
broad, commercially driven audience.
No major PC producer immediately moved to design Internet Protocol
28 The Rise and Stall of the Generative Net
compatibility into its PC operating system. PCs could dial in to a single computer like that of CompuServe or AOL and communicate with it, but the ability to run Internet-aware applications on the PC itself was limited. To attach to
the Internet, one would need a minicomputer or workstation of the sort typically found within university computer science departments—and usually
used with direct network connections rather than modems and phone lines.
A single hobbyist took advantage of PC generativity and forged the missing
technological link. Peter Tattam, an employee in the psychology department of
the University of Tasmania, wrote Trumpet Winsock, a program that allowed
owners of PCs running Microsoft Windows to forge a point-to-point Internet
connection with the dial-up servers run by nascent Internet Service Providers
(ISPs).38With no formal marketing or packaging, Tattam distributed Winsock
as shareware. He asked people to try out the program for free and to send him
$25 if they kept using it beyond a certain tryout period.39
Winsock was a runaway success, and in the mid-1990s it was the primary
way that Windows users could access the Internet. Even before there was wide
public access to an Internet through which to distribute his software, he
claimed hundreds of thousands of registrations for it,40 and many more people
were no doubt using it and declining to register. Consumer accessibility to Internet-enabled applications, coupled with the development of graphic-friendly
World Wide Web protocols and the PC browsers to support them—both initially noncommercial ventures—marked the beginning of the end of proprietary information services and jerry-rigged systems like FIDOnet. Consumers
began to explore the Internet, and those who wanted to reach this group, such
as commercial merchants and advertising-driven content providers, found it
easier to set up outposts there than through the negotiated gates of the proprietary services.
Microsoft bundled the functionality of Winsock with late versions of Windows 95.41 After that, anyone buying a PC could hook up to the Internet instead of only to AOL’s or CompuServe’s walled gardens. Proprietary information services scrambled to reorient their business models away from corralled
content and to ones of accessibility to the wider Internet.42 Network providers
offering a bundle of content along with access increasingly registered their appeal simply as ISPs. They became mere on-ramps to the Internet, with their
users branching out to quickly thriving Internet destinations that had no relationship to the ISP for their programs and services.43 For example, CompuServe’s “Electronic Mall,” an e-commerce service intended as the exclusive
Battle of the Networks 29
means by which outside vendors could sell products to CompuServe subscribers,44 disappeared under the avalanche of individual Web sites selling
goods to anyone with Internet access.
The resulting Internet was a network that no one in particular owned and
that anyone could join. Of course, joining required the acquiescence of at least
one current Internet participant, but if one was turned away at one place, there
were innumerable other points of entry, and commercial ISPs emerged to provide service at commoditized rates.45
The bundled proprietary model, designed expressly for consumer uptake,
had been defeated by the Internet model, designed without consumer demands
in mind. Proprietary services tried to have everything under one roof and to vet
each of their offerings, just as IBM leased its general-purpose computers to its
1960s customers and wholly managed them, tailoring them to those customers’ perceived needs in an ordered way. The Internet had no substantive
offerings at all—but also no meaningful barriers to someone else’s setting up
shop online. It was a model similar to that of the PC, a platform rather than a
fully finished edifice, one open to a set of offerings from anyone who wanted to
code for it.
DESIGN CHOICES AND THE INTERNET ETHOS
Recall that our endpoint devices can possess varying levels of accessibility to
outside coding. Where they are found along that spectrum creates certain basic
trade-offs. A less generative device like an information appliance or a generalpurpose computer managed by a single vendor can work more smoothly because there is only one cook over the stew, and it can be optimized to a particular perceived purpose. But it cannot be easily adapted for new uses. A more
generative device like a PC makes innovation easier and produces a broader
range of applications because the audience of people who can adapt it to new
uses is much greater. Moreover, these devices can at first be simpler because
they can be improved upon later; at the point they leave the factory they do not
have to be complete. That is why the first hobbyist PCs could be so inexpensive: they had only the basics, enough so that others could write software to
make them truly useful. But it is harder to maintain a consistent experience
with such a device because its behavior is then shaped by multiple software authors not acting in concert. Shipping an incomplete device also requires a certain measure of trust: trust that at least some third-party software writers will
write good and useful code, and trust that users of the device will be able to ac30 The Rise and Stall of the Generative Net
cess and sort out the good and useful code from the bad and even potentially
harmful code.
These same trade-offs existed between proprietary services and the Internet,
and Internet design, like its generative PC counterpart, tilted toward the simple
and basic. The Internet’s framers made simplicity a core value—a risky bet with
a high payoff. The bet was risky because a design whose main focus is simplicity may omit elaboration that solves certain foreseeable problems. The simple
design that the Internet’s framers settled upon makes sense only with a set of
principles that go beyond mere engineering. These principles are not obvious
ones—for example, the proprietary networks were not designed with them in
mind—and their power depends on assumptions about people that, even if
true, could change. The most important are what we might label the procrastination principle and the trust-your-neighbor approach.
The procrastination principle rests on the assumption that most problems
confronting a network can be solved later or by others. It says that the network
should not be designed to do anything that can be taken care of by its users. Its
origins can be found in a 1984 paper by Internet architects David Clark, David
Reed, and Jerry Saltzer. In it they coined the notion of an “end-to-end argument” to indicate that most features in a network ought to be implemented at
its computer endpoints—and by those endpoints’ computer programmers—
rather than “in the middle,” taken care of by the network itself, and designed by
the network architects.46 The paper makes a pure engineering argument, explaining that any features not universally useful should not be implemented, in
part because not implementing these features helpfully prevents the generic
network from becoming tilted toward certain uses. Once the network was optimized for one use, they reasoned, it might not easily be put to other uses that
may have different requirements.
The end-to-end argument stands for modularity in network design: it allows
the network nerds, both protocol designers and ISP implementers, to do their
work without giving a thought to network hardware or PC software. More generally, the procrastination principle is an invitation to others to overcome the
network’s shortcomings, and to continue adding to its uses.
Another fundamental assumption, reflected repeatedly in various Internet
design decisions that tilted toward simplicity, is about trust. The people using
this network of networks and configuring its endpoints had to be trusted to be
more or less competent and pure enough at heart that they would not intentionally or negligently disrupt the network. The network’s simplicity meant
that many features found in other networks to keep them secure from fools and
Battle of the Networks 31
knaves would be absent. Banks would be simpler and more efficient if they did
not need vaults for the cash but could instead keep it in accessible bins in plain
view. Our houses would be simpler if we did not have locks on our doors, and
it would be ideal to catch a flight by following an unimpeded path from the airport entrance to the gate—the way access to many trains and buses persists today.
An almost casual trust for the users of secured institutions and systems is
rarely found: banks are designed with robbers in mind. Yet the assumption that
network participants can be trusted, and indeed that they will be participants
rather than customers, infuses the Internet’s design at nearly every level. Anyone can become part of the network so long as any existing member of the network is ready to share access. And once someone is on the network, the network’s design is intended to allow all data to be treated the same way: it can be
sent from anyone to anyone, and it can be in support of any application developed by an outsider.
Two examples illustrate these principles and their trade-offs: the Internet’s
lack of structure to manage personal identity, and its inability to guarantee
transmission speed between two points.
There are lots of reasons for a network to be built to identify the people using it, rather than just the machines found on it. Proprietary networks like
CompuServe and AOL were built just that way. They wanted to offer different
services to different people, and to charge them accordingly, so they ensured
that the very first prompt a user encountered when connecting to the network
was to type in a prearranged user ID and password. No ID, no network access.
This had the added benefit of accountability: anyone engaging in bad behavior
on the network could have access terminated by whoever managed the IDs.
The Internet, however, has no such framework; connectivity is much more
readily shared. User identification is left to individual Internet users and
servers to sort out if they wish to demand credentials of some kind from those
with whom they communicate. For example, a particular Web site might demand that a user create an ID and password in order to gain access to its contents.
This basic design omission has led to the well-documented headaches of
identifying wrongdoers online, from those who swap copyrighted content to
hackers who attack the network itself.47 At best, a source of bad bits might be
traced to a single Internet address. But that address might be shared by more
than one person, or it might represent a mere point of access by someone at yet
another address—a link in a chain of addresses that can recede into the dis32 The Rise and Stall of the Generative Net
tance. Because the user does not have to log in the way he or she would to use a
proprietary service, identity is obscured. Some celebrate this feature. It can be
seen as a bulwark against oppressive governments who wish to monitor their
Internet-surfing populations. As many scholars have explored, whether one is
for or against anonymity online, a design decision bearing on it, made first as
an engineering matter, can end up with major implications for social interaction and regulation.48
Another example of the trade-offs of procrastination and trust can be found
in the Internet’s absence of “quality of service,” a guarantee of bandwidth between one point and another. The Internet was designed as a network of networks—a bucket-brigade partnership in which network neighbors pass along
each other’s packets for perhaps ten, twenty, or even thirty hops between two
points.49 Internet Service Providers might be able to maximize their bandwidth for one or two hops along this path, but the cobbled-together nature of a
typical Internet link from a source all the way to a destination means that there
is no easy way to guarantee speed the whole way through. Too many intermediaries exist in between, and their relationship may be one of a handshake
rather than a contract: “you pass my packets and I’ll pass yours.”50 An endpoint
several hops from a critical network intermediary will have no contract or
arrangement at all with the original sender or the sender’s ISP. The person at the
endpoint must instead rely on falling dominos of trust. The Internet is thus
known as a “best efforts” network, sometimes rephrased as “Send it and pray”
or “Every packet an adventure.”51
The Internet’s protocols thus assume that all packets of data are intended to
be delivered with equal urgency (or perhaps, more accurately, lack of urgency).
This assumption of equality is a fiction because some packets are valuable only
if they can make it to their destination in a timely way. Delay an e-mail by a
minute or two and no one may be the poorer; delay a stream of music too long
and there is an interruption in playback. The network could be built to prioritize a certain data stream on the basis of its sender, its recipient, or the nature of
the stream’s contents. Yet the Internet’s framers and implementers have largely
clung to simplicity, omitting an architecture that would label and then speed
along “special delivery” packets despite the uses it might have and the efficiencies it could achieve. As the backbone grew, it did not seem to matter. Those
with lots of content to share have found ways to stage data “near” its destination
for others, and the network has proved itself remarkably effective even in areas,
like video and audio transmission, in which it initially fell short.52 The future
need not resemble the past, however, and a robust debate exists today about the
Battle of the Networks 33
extent to which ISPs ought to be able to prioritize certain data streams over others by favoring some destinations or particular service providers over others.53
(That debate is joined in a later chapter.)
***
The assumptions made by the Internet’s framers and embedded in the network—that most problems could be solved later and by others, and that those
others themselves would be interested in solving rather than creating problems—arose naturally within the research environment that gave birth to the
Internet. For all the pettiness sometimes associated with academia, there was a
collaborative spirit present in computer science research labs, in part because
the project of designing and implementing a new network—connecting people—can benefit so readily from collaboration.
It is one thing for the Internet to work the way it was designed when deployed among academics whose raison d’être was to build functioning networks. But the network managed an astonishing leap as it continued to work
when expanded into the general populace, one which did not share the worldview that informed the engineers’ designs. Indeed, it not only continued to
work, but experienced spectacular growth in the uses to which it was put. It is
as if the bizarre social and economic configuration of the quasi-anarchist Burning Man festival turned out to function in the middle of a city.54 What works
in a desert is harder to imagine in Manhattan: people crashing on each others’
couches, routinely sharing rides and food, and loosely bartering things of value.
At the turn of the twenty-first century, then, the developed world has found
itself with a wildly generative information technology environment.
Today we enjoy an abundance of PCs hosting routine, if not always-on,
broadband Internet connections.55The generative PC has become intertwined
with the generative Internet, and the brief era during which information appliances and appliancized networks flourished—Brother word processors and
CompuServe—might appear to be an evolutionary dead end.
Those alternatives are not dead. They have been only sleeping. To see why,
we now turn to the next step of the pattern that emerges at each layer of generative technologies: initial success triggers expansion, which is followed by
boundary, one that grows out of the very elements that make that layer appealing. The Internet flourished by beginning in a backwater with few expectations, allowing its architecture to be simple and fluid. The PC had parallel
hobbyist backwater days. Each was first adopted in an ethos of sharing and
tinkering, with profit ancillary, and each was then embraced and greatly im34 The Rise and Stall of the Generative Net
proved by commercial forces. But each is now facing problems that call for
some form of intervention, a tricky situation since intervention is not easy—
and, if undertaken, might ruin the very environment it is trying to save. The
next chapter explains this process at the technological layer: why the status quo
is drawing to a close, confronting us—policy-makers, entrepreneurs, technology providers, and, most importantly, Internet and PC users—with choices we
can no longer ignore.