THE IMPACT OF INTERNET ARCHITECTURE ON THE ECONOMIC ENVIRONMENT FOR APPLICATION INNOVATION

The economic environment for innovation in applications is shaped by two features of the Internet’s original architecture.⁶⁴-⁶⁵ First, the network is general and does not need to be changed to allow new applications to run on the network.

14.5.1 Generality of the Network

In a network based on the broad version of the end-to-end arguments, the network does not have to be changed to allow new applications to run.⁶⁶ First, in such a network, the lower layers of the network - the core of the network⁶⁷ - have been designed to be as general as possible to support a wide variety of current and future applications with different needs, and will therefore be able to support new applications without requiring changes to the network first. Second, the broad version of the end-to-end arguments prohibits applications from implementing application-specific functionality in the core of the network, requiring applications to concentrate their functionality in higher layers at the end hosts instead. Third, since lower layers are independent of higher layers, an isolated change at a higher layer at the end hosts (here: the addition of a new application) will never require changes in the core of the network. In a network based on the broad version, developing a new application thus consists of writing a software program that runs on a computer attached to the Internet, or, more technically, that runs in a higher layer at an end host. All of this affects the economic environment for application innova­tion in a number of ways.

14.4.1.1 Control over application innovation

Since the network does not need to be changed before an application can run on the network, network providers do not have to do anything to enable new applications to run.⁶⁸ As a result, innovators can decide independently whether to realize their idea for an application; they do not need support or ‘permission’ from network providers in order to innovate. This characteristic of the economic environment for innovation is often called ‘innovation without permission’.⁶⁹

14.4.1.2 Cost of application innovation

As developing a new application consists of writing a program that runs on a computer attached to the Internet, innovators can develop new applications at very low cost.⁷⁰ A potential innovator needs access to an end host connected to the Internet, programming knowledge, programming tools, and access to the lower-layer protocols that let the appli­cation send data over the Internet. Like all software programs, the resulting application can be copied and distributed almost cost-free over the Internet.⁷¹ Thus, developing a new application requires almost no physical capital.

14.4.1.3 Size and diversity of the innovator pool

Many of the resources required to develop a new application are often already available (e.g., many people have a computer or access to one) or can be acquired at relatively low cost. As a result, this cost structure lets a wide range of innovators with diverse motiva­tions and funding models develop new applications.⁷² In particular, an innovator does not have to be an employee of a firm or have outside funding to realize his or her idea for an application. Because the biggest investment often is the design and programming of the application itself, potential innovators can develop an application in their free time or as a ‘side project’, with the opportunity cost of the time spent the most important cost factor. Under these conditions, an application does not have to produce a profit in the future to cover the costs of developing it.

14.4.1.4 Control over application deployment and use

Before a user can use a new application, he or she needs to install it on their computer. Due to the generality of the network’s core, deploying or using a new application does not require any changes to the core.⁷⁵ Thus, network providers do not have to do anything before an application can run on the network. As a result, each user can decide independ­ently which applications he or she wants to deploy and use.⁷⁶-⁷⁷ This characteristic of the economic environment for innovation is often called ‘user choice’.

14.4.2 Application Blindness of the Network

Like all networks that comply with the broad version of the end-to-end argument, the original Internet was application-blind, that is, it was unable to distinguish among the applications on the network. As a result, it was unable to make distinctions among data packets based on that information, for example, when transporting them across the network or when charging for them.

The application blindness of the network deprived network providers of three strate­gic options that would be available in an application-aware network.⁷⁸ First, a network provider in an application-aware network can block applications or discriminate against them. Second, it can charge an access fee to providers of applications and content who are not the network providers’ Internet service customers. Third, it can charge different Internet transport prices for different applications, or it can exclude applications to price discriminate between customers of its Internet service.

14.4.2.1 Blocking or discrimination

A network provider in an application-aware network can identify the applications on its network and can control their execution. This enables network providers to block appli­cations or to discriminate against them. If a network provider blocks an application, the network provider’s Internet service customers will be unable to use it. Discrimination is differential treatment that falls short of blocking, but makes an application or class of applications relatively less attractive.⁷⁹ For example, a network provider may slow down a specific application, or speed up competing applications. Blocking or discrimination allows network providers to distort competition among applications or classes of appli­cations, and to interfere with the decisions of application innovators and users regarding application innovation, deployment, and use. Thus, an application-aware network shifts control over application innovation, deployment, and use to network providers.⁸⁰

By contrast, since an application-blind network is unable to distinguish among applications, a network provider in such a network is unable to block applications or discriminate against them. If the network is application-blind and general, control over application innovation rests with application innovators, while users control application deployment and use.⁸¹-⁸²

Blocking or discrimination enables network providers to distort competition among applications or classes of applications. An innovator whose application is blocked will be unable to benefit from the application. Discrimination puts an application at a competitive disadvantage relative to other applications that are not discriminated against, reducing the benefits an innovator can expect to realize. Thus, the threat of discrimination reduces the expected benefits of innovation, which reduces innovators’ incentives to develop new applications and their ability to get funding.⁸³ As statements from entrepreneurs show, the threat of discrimination in the absence of network neu­trality rules reduces innovators’ ability to get funding today, so this is not just a theoreti­cal concern.⁸⁴

Network providers’ ability to block or discriminate against applications can only affect application innovation, however, if network providers have an incentive to block or discriminate.

Over the past ten years, all of these types of discrimination have occurred in prac­tice. For example, network providers in the USA, Europe, Saudi Arabia, United Arab Emirates and Mexico blocked Internet telephony applications such as Skype because the use of Internet telephony reduces network providers’ revenue from traditional telephony service.⁹⁰ Cable providers in the USA have expressed their desire to interfere with over-the-top online video applications such as Netflix to protect their conventional television offerings.⁹¹ Network providers in the USA prohibited the use of virtual private network applications on their basic Internet service offerings to price discriminate among their Internet service customers.⁹² Network providers in the USA, Canada and Europe singled out specific applications or classes of applications to manage bandwidth on their networks.⁹³ A network provider in Canada blocked content that threatened its business interests, while network providers in the Middle East, China and other countries blocked content deemed politically, religiously, socially, or culturally inappropriate.⁹⁴ In many of these cases, the network provider did not participate in the market for the excluded appli­cation, and when it did participate, the exclusion did not allow the network provider to monopolize the market for the application in question.

Commentators often assume that competition in the market for Internet services will remove any incentives to engage in blocking or discrimination.⁹⁵-⁹⁶ If there is competi­tion and a network provider discriminates against an application that users would like to use, they argue, users can switch to another network provider that does not discriminate against the application, and this threat of switching will discipline providers.

These arguments fail to recognize that the market for Internet service is characterized by incomplete customer information, product differentiation in the market for Internet access and for wireless and wireline bundles, switching costs, and, in some countries, a concentrated market structure in the market for Internet services. These factors limit the effectiveness of competition, even in markets with several competing Internet service pro­viders, and reduce consumers’ willingness to switch Internet service providers in response to discriminatory conduct, giving network providers a degree of market power that enables them to impose restrictions on their Internet service customers that they would not be able to impose in a perfectly competitive market.⁹⁷

In line with this theoretical argument, network providers in markets that are more competitive than the market for wireline, fixed Internet service in the USA have engaged in blocking or discrimination.⁹⁸ This evidence suggests that at least in the market for wire­line Internet service in Europe and Canada and in the market for mobile Internet service in the USA, competition does not prevent Internet service providers from interfering with applications, content, or services on their networks, even if, as in the USA and in the European Union, network providers are required to disclose any discriminatory conduct that occurs.⁹⁹

14.4.2.2 Access fees

A network provider in an application-aware network can charge an access fee to provid­ers of applications and content who are not its Internet service customers.¹⁰⁰ By contrast, a network provider in a network based on the broad version is unable to do so, since it cannot distinguish among the applications on its network.

Access fees come in two variants.¹⁰¹ In the first, a network provider charges applica­tion providers who are not its Internet service customers a fee for the right to access the network providers’ Internet service customers. Applications whose providers do not pay the access fee cannot be used on the network provider’s access network. In the second variant, a network provider charges application providers for prioritized or otherwise enhanced access to the network provider’s Internet service customers. For example, if an application provider has paid such an access fee, the application’s data packets may receive a better type of service (e.g., travel faster) on the network provider’s access network or may not count against a user’s monthly data cap.¹⁰²

Access fees increase the costs of offering an application for all application providers subject to the fee, reducing the profits they can expect to make. The theory of two-sided markets even predicts that network providers would charge monopoly prices to applica­tion providers, which would reduce application providers’ incentives to innovate even further.¹⁰³ In addition, access fees will disproportionately affect innovators with little or no outside funding: innovators with little or no outside funding may not be able to pay access fees, which would put them at a competitive disadvantage to those who can and do pay the fees.¹⁰⁴ Innovators who cannot pay the first type of access fees will be unable to compete at all. Those who cannot pay the second type of access fees will be able to reach end users, but their applications will perform worse or will cost users more (e.g., because the use of the application counts against the user’s monthly data cap) than the applica­tions of providers that pay the access fee. Thus, access fees will reduce the incentives of innovators with little or no outside funding to innovate and may remove (or at least impede) their ability to innovate in offerings subject to access fees, which in turn would reduce the size and diversity of the pool of potential innovators.¹⁰⁵ Finally, network pro­viders can influence which applications are successful by allowing only selected applica­tion providers to pay for prioritized or otherwise enhanced treatment.¹⁰⁶

14.4.2.3 Application-specific pricing and price discrimination

In an application-aware network, a network provider can identify which applications a user uses and can control their execution. This enables network providers to charge differ­ent Internet transport prices for different applications (e.g., charge higher Internet service fees for an e-mail packet than for a packet of web content of equal size), or to price dis­criminate among customers of its Internet service by excluding applications (e.g., allow the use of video conferencing only for users of its premium Internet service, not for users of its basic Internet service). By contrast, a network provider in an application-blind network can base Internet transport prices only on observable characteristics such as the total amount of bandwidth used by an Internet service customer, not on the application being used. Thus, network providers in an application-blind network have to charge the same quality- and bandwidth-adjusted Internet transport price to all applications.¹⁰⁷

A network provider that can charge application-specific transport prices or price dis­criminate by excluding applications will be able to extract more of the consumer surplus associated with a specific application through the pricing of its Internet service than network providers in an application-blind network, leaving less consumer surplus and, therefore, less potential profits for the provider of that application and, in the case of price discrimination, for application providers in general.

14.4.3 Impact of Architectural Changes on the Amount and Quality of Application Innovation

In sum, the broad version of the end-to-end arguments created an economic environment for application innovation that is characterized by innovation without permission, user choice, application blindness, and low costs of application innovation (see Table 14.1). Together, these elements of the economic environment for innovation affect the amount and quality of application innovation that will occur under an architecture based on that design principle.

Architectures that deviate from the broad version of the end-to-end argument will result in different economic environments for application innovation. Architectures can deviate from the broad version along two dimensions, and, along each of these dimen­sions, to varying degrees: relative to an architecture based on the broad version, they can become more opaque, more controllable, or both.¹⁰⁸

First, network providers can reduce the generality of the network by implementing, in the network’s core, application-specific functionality that optimizes the network for the needs of existing applications. These changes make an architecture more opaque. Changes that make the network more opaque are often driven by the desire to increase performance or save costs.¹⁰⁹

Second, network providers can remove the network’s application blindness by adding, in the network’s core, application-specific functionality that increases network providers’ ability to monitor and control the applications on their networks. These changes make

Table 14.1 The economic environment for innovation under different network architectures

an architecture more controllable. Network providers primarily move in this direction to increase their profits; however, changes that let network providers monitor what is hap­pening on the network may also help them make the network more secure, manage traffic, or plan network upgrades.¹¹⁰

Over the past two decades, the Internet’s architecture has become more opaque and more controllable. The past two decades have seen a proliferation of technical solutions that reduce the generality of the network by optimizing the network for the needs of a particular class of applications called client-server applications. For example, residential broadband networks provide asymmetric bandwidth for uploads and downloads, creat­ing problems for applications that send and receive equal amounts of data.¹¹¹ To deal with the impeding shortage of Internet addresses, users and network providers deployed network-address translators that let several devices share the same Internet address. To protect their networks against attacks, organizations put firewalls that block potentially harmful applications at the borders of their private networks. Network-address trans­lators and firewalls have made it increasingly difficult to deploy new transport-layer protocols or new applications whose behavior deviates from that of traditional client­server applications.¹¹² At the same time, technologies such as deep packet inspection have removed the application blindness of the network. They provide network providers with fine-grained awareness of and control over the applications on their networks, and allow them to implement the various pricing schemes (e.g., application-specific pricing or access fees) discussed throughout this chapter. These technologies have been widely deployed.¹¹³

How exactly deviations from the broad version of the end-to-end arguments affect the economic environment for application innovation depends on the exact nature of the architectural change.¹¹⁴ Most generally, deviating from the broad version will increase the costs or reduce the benefits of application innovation, result in an innovator pool that is smaller or less diverse, or shift control over application development, deployment, and use to network providers. In the worst case, an architectural change has all of these effects (see Table 14.1).

Other things being equal (such as the group of actors exposed to the architecture or the other constraints operating on the actors) and assuming the absence of legal rules that limit network providers’ ability to take advantage of the capabilities provided by the architecture, differences in these characteristics of the economic environment for innova­tion will affect the amount and quality of application innovation that will occur.

14.4.3.1 Costs and benefits of innovation

Some architectural changes may affect the benefits and costs of application innovation. An innovator decides to innovate if the expected benefits (broadly defined) are larger than the costs. If an architectural change increases the costs or reduces the expected benefits of an innovation relative to another architecture, some innovations that were justified under that other architecture may not be justified any more under the changed architecture.¹¹⁵ For example, relative to an application-blind architecture, an application- aware architecture that allows network providers to block applications, charge access fees, or charge application-specific prices reduces the benefits an application developer can expect to reap.¹¹⁶ Increasing the amount of application-specific functionality in the core of the network increases the costs of application innovation.¹¹⁷ Thus, under these architectures, the same group of innovators will find less of their innovative ideas for applications worth pursuing than under an architecture based on the broad version.¹¹⁸

In addition to reducing the expected profitability of application innovation, an increase in the costs of application innovation or deployment may affect innovators’ ability to serve niche markets or low-value markets, and may reduce their willingness to take risks.¹¹⁹

14.4.3.2 Size and diversity of the innovator pool

Architectural changes may also influence who can innovate, affecting the size and diver­sity of the pool of potential innovators.¹²⁰

Under an architecture based on the broad version of the end-to-end arguments, almost anybody can be an innovator. Developing a new application requires only access to an end host connected to the Internet, programming knowledge, programming tools, and access to the lower-layer protocols that let the application send data over the Internet. This allows innovators with a wide range of motivations and funding models to develop new applications.¹²¹

By contrast, architectures that deviate from the broad version increase the requirements that potential innovators must meet, which reduces the size and diversity of the innovator pool.¹²² For example, in an application-aware architecture that allows network providers to charge access fees, innovators with little or no outside funding will often be unable to pay access fees, which reduces or removes their ability to develop new applications.¹²³ Similarly, if the core of an architecture contains application-specific functionality, some applications may require changes to the network’s core before they can run on the network.¹²⁴ As a result, the costs of developing and testing these applications will be significantly higher than in a network based on the broad version. They include not only the costs of developing and testing the application itself, but also the costs of developing and testing the changes to the network’s core and the subsequent changes to applications that relied on the former version of the network’s core. Because the application cannot run on the operational network, the innovator must have access to a test network. If the changes to the network’s core include changes to hardware, the new hardware must be manufactured and physically distributed. Thus, the costs of production and distribution also may be higher under such architectures. Finally, the developer may incur consider­able costs while trying to convince network providers of the usefulness and commercial viability of the new application and negotiate any necessary changes to the network’s core and to other existing applications. Under such an architecture, many of the smaller or non-commercial innovators that would have been able to develop applications under an end-to-end architecture will lack the resources and capabilities necessary to over­come these technical, financial, or coordinational hurdles. Depending on the amount of application-specific functionality in the network’s core, these obstacles may be so severe that independent innovators other than network providers lose the ability to innovate.¹²⁵

If there is uncertainty (e.g., about technology or user needs) or if user needs are het­erogeneous, differences in the size and diversity of the innovator pool will affect the amount and quality of application innovation under the different architectures.¹²⁶ If there is uncertainty, nobody really knows which applications will work, or which applications will be successful. Under these circumstances, evolutionary and neo-institutional theories of innovation predict that a larger, more diverse group of innovators will not only iden­tify a larger, more diverse set of opportunities for innovation than a smaller, less diverse group; they will also realize a larger number of the opportunities that were discovered.¹²⁷ As I have shown in detail elsewhere, the history of application innovation on the Internet strongly supports these theories.¹²⁸

Discovering opportunities for innovation is the first step in the innovative process.¹²⁹ Often it is not immediately apparent how a new technology could be used, or that cus­tomers have unmet needs that a new or improved product could fill. This problem is particularly pronounced for relatively general technologies, such as the Internet, that can be put to a wide variety of uses. For these technologies, identifying ways in which the technology could be used is an important kind of innovation. As history has shown again and again, it is usually impossible to decide in advance which new uses will become successful.¹³⁰ Most Internet applications that later became highly successful either were not envisaged by the designers of the network or were met by widespread skepticism when they first became available. This was true, for example, for e-mail, the World Wide Web, eBay, or search engines.¹³¹

Different actors have different motivations, backgrounds, and theories about the world. As a result, they will see different opportunities for innovation.¹³² Even if different actors recognize the same opportunity, they may differ in their perception of the likeli­hood of success or of the likely costs and benefits. As a result, the same opportunity may appear profitable to some of the innovators, but not to others.¹³³ The impact of these differences among actors increases with uncertainty.¹³⁴ Beyond differences in perception, differences in actual cost structures may make a project profitable for some innovators but not for others.¹³⁵ For example, an innovator working from home in his or her free time incurs lower costs than a start-up that rents offices and pays its employees. That start-up, in turn, will often have lower costs than a large corporation with significant overhead. And even if the expected benefits of the application are larger than the costs for a number of innovators, the size of the potential benefits may be unattractive for some potential innovators, but not for others.¹³⁶ For example, large companies are interested in projects that have the potential to contribute significantly to their bottom line. Venture capitalists invest in projects that have the potential for extremely large rewards, because the profits from a few successful projects need to cover the venture capitalists’ losses on their other projects. By contrast, entrepreneurs who finance a project themselves or through angel investors may be willing to invest in projects that large corporations or venture capital funds may not find attractive. Similarly, a user who wants to use the application them­selves will not care about the size of the market at all, since he or she is motivated by the prospect of being able to use the application themselves.

For all these reasons, a larger, more diverse group of innovators will identify and realize a larger, more diverse set of potential projects than a smaller, less diverse group. Having a larger and more diverse set of innovation projects, in turn, guarantees a more complete search of the problem space, increasing the chance that beneficial uses (or approaches to specific problems) will be detected. If the various innovators pursue different technical approaches to the same problems or explore different business models, having a larger number of innovators increases the likelihood that at least one of them will be successful and increases the expected quality of the best results.¹³⁷ If users’ needs are heterogeneous, a larger and more diverse group of potential innovators will also create a greater variety of products that better meet users’ needs.¹³⁸ Thus, if there is uncertainty or user needs are heterogeneous, an increase in the size and diversity of the innovator pool will increase the amount and quality of application innovation.

Diversity is not without costs. Each approach incurs costs, and many projects will fail. Under fundamental uncertainty, successful projects cannot be identified in advance, so these costs seem to be unavoidable.¹³⁹

14.4.3.3 Control over application innovation

Under an architecture based on the broad version of the end-to-end arguments, innova­tors independently choose which applications they want to pursue, without interference from network providers. Due to the generality of the network’s core, they do not need support or ‘permission’ from network providers in order to realize their idea for an appli­cation (‘innovation without permission’), and the application blindness of the network prevents network providers from interfering with these choices.¹⁴⁰

By contrast, under architectures that deviate from the broad version, network pro­viders may need to take action before an application can be realized. For example, if a network contains application-specific functionality in the network’s core, the network may need to be changed before a new application can function on the network.¹⁴¹ In an application-aware architecture, the network can be closed to new applications by default. In such a network, the network provider needs to enable an application before it can run on the network.¹⁴² Depending on the architecture, contracting with a network provider may be the only way for potential innovators to realize their idea for an application. This may happen, for example, if the network’s core contains a significant amount of application-specific functionality.¹⁴³

The difference in control over application innovation affects the amount of innovation in various ways.¹⁴⁴ First, if there is uncertainty, increasing the number of decision-makers who need to approve an idea or take action before an idea can be realized reduces the chances that the idea will be realized. A network provider will see no need to support applications that it perceives as infeasible, as not viable, or as counter to its strategic inter­ests. These projects would be realized under an end-to-end architecture, but not under an architecture in which network providers need to approve new applications.

Second, an innovator has to disclose its innovation when negotiating for the network provider’s cooperation. If the innovation is not protected by intellectual property rights, there is a danger that the network provider may appropriate the innovation without paying for it, which may reduce the innovator’s incentive to innovate in the first place.

Third, bargaining costs and strategic behavior may prevent the innovator and the network provider from reaching agreement.

Fourth, the incentives for independent innovators who can directly commercialize their innovation are higher owing to the possibility of exceptionally large gains. If innovators have to contract with a network provider in order to commercialize their innovation and cannot gain access to the market on their own, they do not have any bargaining power. In such a network, the network provider will be in a monopsony or oligopsony position, which leads to very low prices. In this case, the innovators will have to bear the risk of failure or bargaining breakdown, and will receive only modest compensation if they succeed. Such an incentive structure will probably not be sufficient to motivate innova­tors or their investors to put up with the risk.

Finally, if innovators have to contract or otherwise coordinate with network providers before they can innovate, they may be less able to react to new developments once they have started their project. Throughout the history of the Internet, successful innovators have often changed course repeatedly - for example, in response to feedback from consumers, or in response to an unexpected scarcity of funding. This happened, for example, in the cases of Blogger, Flickr, or PayPal. More generally, starting with one approach and then adapt­ing it as events unfold may be the only way to successfully navigate fundamental uncer­tainty. In a network that requires innovators to coordinate their activities with the network provider, this may not be possible. According to transaction-cost and coordination-cost theories, deciding what to do in response to new developments is much more difficult and more time-consuming across firm boundaries than within a single firm. Firms have dif­fering perspectives on how to react, and the lack of efficient mechanisms for inter-firm decision-making and dispute resolution may make the differences difficult to resolve.

Thus, relative to architectures in which innovators can innovate independently, archi­tectures that require innovators to get approval from, contract with, or otherwise coordi­nate with a network provider before they can innovate constrain independent innovators’ ability and incentives to start a project and reduce the chance that innovative ideas will be realized. They also limit innovators’ ability to react to new developments during the lifetime of the project - a limitation that is particularly detrimental under uncertainty.

14.4.3.4 Control over application deployment and use

Under an architecture based on the broad version of the end-to-end arguments in which users control the end hosts,¹⁴⁵ users independently decide which applications they want to deploy and use. This characteristic has been called ‘user choice’. Due to the general­ity of the network, network providers are not involved in the deployment and use of applications, and the application blindness of the network’s core prevents them from interfering with users’ choices regarding application deployment and use.¹⁴⁶

By contrast, architectures that deviate from the broad version often shift control over application deployment or use to network providers. For example, if a network needs to be changed before an application can run on the network (e.g., because an application- aware network enables network providers to block new applications by default or because a network contains application-specific functionality that prevents the application from running), the network provider controls which applications can get deployed. If an application-aware network allows network providers to block applications or discrimi­nate against them, network providers can control how the network is used.¹⁴⁷

Allowing users to independently choose which applications to deploy and use increases the chance that a new application will be deployed and used. In a network based on the broad version, the only person who needs to be convinced that an application may be useful for the application to be deployed is the person who actually wants to use it. By contrast, increasing the number of decision-makers who need to take action or approve an application before it can be deployed reduces the chance that the application will be deployed.¹⁴⁸

In addition, letting users choose which applications to deploy and use is an important part of the mechanism that produces innovation under uncertainty. If there is uncer­tainty, nobody knows in advance which applications will work, or which applications will be successful. Under these circumstances, neo-institutional and evolutionary theories of innovation suggest that it is best to try out many different ideas, and see what happens. Some applications may succeed, and some may fail, but trying is the only way to find out.¹⁴⁹ In a network that allows for this kind of experimentation, the actors that control application deployment and use effectively decide which applications will be successful.¹⁵⁰

When choosing which applications will be successful, users and network providers will often make different decisions. Thus, differences in control over application deployment and use will result in different applications being successful.

First, network providers’ interests may differ from users’ interests, which can lead them to reject applications that users may find attractive.¹⁵¹ Network providers and users use different criteria when choosing which application should be deployed and used. Users choose the applications that best meet their needs. By contrast, network providers will choose the applications that maximize their profits. As a result, they may, for example, reject applications that compete with their own applications or with a partner’s applica­tion, block certain applications in order to price discriminate among their Internet service customers, exclude content that threatens their business interests or does not comply with their content policies, or single out applications or classes of applications to manage bandwidth on their network.¹⁵² The market for Internet services is afflicted with a number of factors - incomplete customer information, product differentiation in the market for Internet access and for wireline and wireless bundles, and switching costs - that provide network providers with market power, even in markets with several competing Internet service providers. This market power allows them to impose restrictions on their users that they would not be able to impose in a perfectly competitive market, enabling them to exclude applications that users want.¹⁵³

Second, even when network providers would like to deploy applications that users want, they do not necessarily know what these are. New applications are often afflicted with con­siderable uncertainty. No one knows which applications or features users will find attrac­tive. Often users themselves do not know whether they like a specific application or find it

useful until they have tried it or seen others using it. Thus, letting users choose applications replaces network providers’ guesses about what users may like with a decision by users, who best know their current preferences and needs. It provides a chance to applications whose usefulness or attractiveness is not immediately apparent, and allows users’ preferences and practices to evolve as a result of exposure to or experimentation with a new application. In addition, giving application developers direct access to users lets them experiment and change their product offerings in response to user feedback, which has been critical for applications like Flickr, Blogger, or PayPal. Finally, being able to choose which applica­tions to deploy and use is important for users or user groups whose needs are idiosyncratic, since their needs may not be known to or may not be important to the network provider.¹⁵⁴

Beyond innovation, letting users choose how they want to use the Internet enables them to use the Internet in a way that creates more value for them and for society than if network providers made this choice.¹⁵⁵

In sum, the mechanism that produces application innovation under an architecture based on the broad version of the end-to-end arguments has two components: first, widespread experimentation by a large and diverse group of innovators who can innovate at low cost and who independently select whether to realize their innovative ideas, and second, user choice among the resulting applications. Under uncertainty or consumer heterogeneity, this mechanism will produce more and better applications than innova­tion in network architectures that reduce the size and diversity of the pool of potential innovators or concentrate control over innovation, deployment, or use in the hands of network providers.¹⁵⁶ If user needs are heterogeneous, this mechanism will also produce more diverse applications that better meet user needs.¹⁵⁷

In the current Internet, technological uncertainty, market uncertainty, and user het­erogeneity are high, so the conditions under which innovator diversity, user choice, and innovation without permission increase the amount and quality of application innova­tion are met.¹⁵⁸

14.6 CONCLUSION

Network architectures (and the design principles that shape them) influence the eco­nomic system for application innovation, deployment, and use in different ways. In par­ticular, they influence which actors can innovate, what incentives they have to do so, and who controls whether an application can be developed, deployed, and used.

The Internet’s original architecture was based on the layering principle and the broad version of the end-to-end arguments. This design created an economic environment for application innovation that is more conducive to innovation in applications than archi­tectures that deviate from this design principle.

In networks based on the broad version of the end-to-end arguments, the economic environment for application innovation is characterized by application blindness, inno­vation without permission, user choice, and low costs of application innovation. Architectures that deviate from the broad version will result in different economic environments for application innovation. How exactly deviations from the broad version of the end-to-end arguments affect the economic environment for application innova­tion depends on the exact nature of the architectural change. Most generally, deviating from the broad version will increase the costs or reduce the expected benefits of appli­cation innovation, reduce the size and diversity of the innovator pool, or shift control over application innovation, deployment, or use from application innovators and users to network providers. In the worst case, an architectural change has all of these effects. Other things such as the set of actors exposed to the architecture and the set of con­straints under which they operate all being equal, these kinds of changes will reduce the overall amount, the type, and the quality of application innovation. If there is market uncertainty, technical uncertainty, or user heterogeneity, the effects of differences in the size and diversity of the innovator pool and of differences in control over application development, deployment, and use are particularly profound. In the current Internet, there is uncertainty and user needs are heterogeneous, so the conditions under which innovator diversity, innovator control over application innovation, and user control over application deployment and use increase the amount and quality of innovation are met.

To highlight the specific effect of architecture, this chapter assumed that while the architecture changes, everything else (i.e., the group of actors exposed to the architec­ture and the other constraints under which they operate) stays the same. In reality, the actual effect of a specific architecture or architectural change on innovation at a specific place and time cannot be determined without considering the characteristics of the actors exposed to the architecture, the other constraints (such as laws, norms, and the natural and technical environment) under which they operate, and the actors’ existing or expected relationships with others.

For example, while the broad version of the end-to-end arguments allows anybody with access to an end host, programming tools, and programming knowledge to develop new applications, factors like an entrepreneurial culture that embraces the possibility of failure or the availability of sources of financing like angel investors and venture capital­ists that specialize in early-stage, high-risk projects may make it more likely that a poten­tial innovator with an idea for an application decides to realize it.

Similarly, an innovator’s ability to leverage the opportunities provided by the architec­ture into a successful innovation may be influenced by the organizations, cultures, and networks in which he or she is embedded. For example, Yahoo! and Google started at Stanford University, a prestigious and well-connected institution with many formal and informal connections to Silicon Valley and with an infrastructure that encourages and supports student-led and faculty-led entrepreneurship. Other innovators with the same idea and similar personal resources who faced the same architectural constraints, but who were embedded in different organizations, cultures, or networks, may have been less suc­cessful in realizing the same innovation.¹⁵⁹

Finally, the effect of architecture may be mediated by factors such as the amount of competition in the market for Internet services or the existence of laws and regulations that limit network providers’ ability to the use the capabilities provided by an architecture (‘network neutrality rules’). For example, a network non-discrimination rule that prevents network providers from discriminating among applications or classes of applications based on application-specific criteria, but allows application-agnostic discrimination, would recreate through law the same economic environment for application innovation and network use with respect to discrimination as an application-blind network, regard­less of whether the underlying architecture is application-aware or not.¹⁶⁰-¹⁶¹ By contrast, in the absence of rules limiting discrimination, application-blind and application-aware architectures result in markedly different environments for application innovation and network use.

Thus, while the architecture of a network affects the economic environment for application innovation, deployment, and use in important ways, it is only one factor in a complex system that jointly influences the overall amount and quality of application innovation that will occur. Therefore, although the Internet’s architecture in most coun­tries has generally been the same, different countries have seen and will likely see different amounts of application innovation.

NOTES

* Parts of the chapter are adopted from van Schewick (2010a). Permission by MIT Press to reprint excerpts from the book is gratefully acknowledged. Due to space constraints, notes are reprinted only selectively. For a more detailed analysis of the questions discussed in this chapter and the full set of refer­ences, see van Schewick (2010a). For shorter summaries, see van Schewick (2010c; 2015a, pp. 19-23).

↑ Professor of Law and Helen L. Crocker Faculty Scholar, Stanford Law School; Director, Center for Internet and Society, Stanford Law School; Professor (by Courtesy) of Electrical Engineering, Stanford University. The author would like to thank Elaine Adolfo and Dolfin Leung for creating the figures and the librarians at Stanford Law School for their incredible work.

1. Throughout this chapter, the term original architecture of the Internet refers to the network architecture that was specified in the DARPA Internet Program protocol specifications for the Internet Protocol and the Transmission Control Protocol, RFC 791 (Postel, 1981a) and RFC 793 (Postel, 1981b). David Clark described this architecture in an important article on the design philosophy of the DARPA Internet Protocols (Clark, 1988).

2. Throughout this chapter, the term innovation refers to creating or improving goods, services, or methods of production (e.g., Hall, 1994, p. 2; Beije, 1998, pp. 1-2). As used here, the term innovation is not restricted to innovative activities that firms undertake to increase their profits (for this more restrictive use of the term, see, e.g., Hall, 1994, pp. 17-19), but denotes innovative activities by a wide range of actors - including individuals, groups, firms, and other organizations - whose actions and products may be motivated by various economic or non-economic concerns (van Schewick, 2010a, p. 28).

3. For a discussion of this term and more complete references to the literature exploring the economic effect of technical architectures, see van Schewick (2010a, pp. 13-14). Since the 1990s, scholars of management strategy have explored the economic effect of modular and integrated architectures. See, for example, Langlois and Robertson (1992); Garud and Kumaraswamy (1993); Ulrich (1995); Sanchez (1995); Baldwin and Clark (2000). The effect of the Internet’s original architecture on innovation was first high­lighted by Isenberg (1997); Isenberg (1998); Reed et al. (1998); Lemley and Lessig (1999); Lessig (2001).

4. See, for example, van Schewick (2010a, pp. 106-7, 151-4, 157-63, 366, 382-3 discussing the impact of Internet architecture on innovation in the Internet layer and pp. 88-90, 148-55 discussing the impact of Internet architecture on innovation in the link layer).

5. The analysis does not distinguish among Internet service providers that are vertically integrated into the operation of the network infrastructure and those that provide Internet access or transport services over another provider’s physical infrastructure. See also van Schewick (2010a, p. 222).

6. van Schewick (2010a, pp. 20-21).

7. The framework described in the text is used by researchers in economics (North, 1990; Aoki, 2001; Furubotn and Richter, 2005; Greif, 2006), political science (Hall and Taylor, 1996; Thelen, 1999; Peters, 2005); law (Sunstein, 1996; Lessig, 1998; Benkler, 2006), and sociology (Powell and DiMaggio, 1991; DiMaggio, 1998; Scott, 2000). For a more detailed description of this framework with references to the literature, see van Schewick (2010a, pp. 23-8).

8. van Schewick (2010a, pp. 4, 28-30).

9. The effect of the economic system on the evolution of architectures is outside the scope of this chapter. On that topic, see, for example, van Schewick (2010a, pp. 3, 23-6, 28, 32, 151-63, 371-2, 389-402).

10. van Schewick (2010a, pp. 30-32).

11. van Schewick (2010a, p. 30).

12. van Schewick (2010a, p. 30). Economists usually treat resources and cost structures as constraints. Since they are specific to the individual actor, I mention them here. See also Ostrom (2005, pp. 828-9); Furubotn and Richter (2005, pp. 3, 308-9).

13. van Schewick (2010a, p. 31).

14. See, for example, Landes and Posner (2003).

15. Lessig (1998, pp. 663-4); van Schewick (2010a, pp. 26-8, 32).

16. For an example, see the Conclusion.

17. van Schewick (2010a, p. 30). This part of the framework also allows the integration of insights from sociology about the effect of existing social or economic relationships on economic behavior. See, for example, the research on embeddedness (e.g., Granovetter, 1985; Uzzi, 1996) or on the importance of organizational fields for firms’ organizational choices (e.g., DiMaggio and Powell, 1983). On the rel­evance of embeddedness within organizations, cultures, or network connections for innovation, see, for example, Saxenian (1994); Castilla et al. (2000); Powell and Grodal (2005). See also the Conclusion.

18. See, for example, Dyer (1996, 1997); Takeishi (2001); van Schewick (2010a, pp. 178-9, 190).

19. van Schewick (2010a, pp. 32-3).

20. On the impact of such rules, see the Conclusion. See also van Schewick (2010a, pp. 218-21).

21. van Schewick (2010a, pp. 19-21).

22. van Schewick (2010a, pp. 50-52).

23. This description glosses over the fact that in some network architectures, the end hosts may not only use, but also offer network services. In the Internet, the end hosts (or computers ‘on’ the network) also par­ticipate in the operation of the network through the protocols at the Internet layer and below. Thus, one may say that in the Internet, end hosts form the network too. A more precise description of the Internet would focus on layers: the layers up to the Internet layer form or implement the network and are ‘in’ the network, while the layers above the Internet layer use the network and are ‘on’ the network. See, for example, Comer (2000, p. 186); Sterbenz and Touch (2001, p. 350).

24. Throughout this chapter, the term the core of the network will be used to denote the set of computers in the network, or, in the case of the Internet, the lower layers up to, and including, the Internet layer (see note 23). For a similar use, see Blumenthal and Clark (2001, pp. 71-2). Sometimes (but not in this chapter), the term core network is used to denote the part of a hierarchical telecommunications network that provides the highest level of aggregation, such as the backbone network, as opposed to the inter­mediate part of the entire network, which connects the core network with the access networks or edge networks.

25. van Schewick (2010a, pp. 50-51, 378, 107-10).

26. Layering, in turn, is a special form of modularity (e.g., van Schewick, 2010a, p. 46). On modularity, see van Schewick (2010a, pp. 38-44).

27. For a detailed discussion of the layering principle in the context of networking with references to the literature, see van Schewick (2010a, pp. 50-57). See also Tanenbaum and Wetherall (2011, pp. 29-33); Peterson and Davie (2012, pp. 24-31). On the layering principle in general, see van Schewick (2010a, pp. 46-50).

28. van Schewick (2010a, p. 88).

29. van Schewick (2010a, pp. 57-8). With respect to the International Organization for Standardization’s Open Systems Interconnection reference model (the transport layer and higher layers are typically imple­mented on end hosts, not on the intermediate switches or routers), see, for example, Peterson and Davie (2012, pp. 32-3); Sterbenz and Touch (2001, pp. 41-2). With respect to the architecture of the Internet (end hosts implement all layers, while IP routers typically implement only lower layers, up to and includ­ing the Internet layer), see Kurose and Ross (2010, pp. 54-5). In practice, routers may implement higher layers to terminate routing protocols such as BGP or management protocols.

30. The text describes the ‘relaxed’ version of the layering principle. In the pure version of layering, a layer is allowed to use only the layer immediately below it. On the various versions of the layering principle, see van Schewick (2010a, pp. 46-7).

31. On the benefits and costs of layered architectures, see van Schewick (2010a, pp. 47-9).

32. See, for example, van Schewick (2010a, pp. 51-2, 54 6); Comer (2000, pp. 187-9); Sterbenz and Touch (2001, pp. 42-3).

33. See, for example, van Schewick (2010a, pp. 51-2, 56); Reed (2010); Peterson and Davie (2012, pp. 30-31).

34. See note 48 and accompanying text. In addition to being application-blind, the resulting network is also application-agnostic. While an application-agnostic network may have information about the applica­tions on the network, it does not make distinctions among data packets based on that information. Thus, an application-blind network is necessarily application-agnostic: it does not make distinctions among data packets based on information about the applications on the network, because it does not have this information. By contrast, an application-agnostic network is not necessarily application-blind, because it may have information about the applications on the network. See van Schewick (2015a, pp. 24-6, Box 4).

35. van Schewick (2010a, pp. 57-8).

36. Saltzer et al. (1981). The 1981 paper was a conference paper. When referring to the original paper that identified the end-to-end arguments, researchers usually refer to the revised version (Saltzer et al., 1984) that appeared in the ACM Transactions on Computer Systems.

37. See, for example, Reed et al. (1998); Reed (2000); Blumenthal and Clark (2001). See also Clark and Blumenthal (2011, pp. 383-8) (tracing the evolution of the end-to-end arguments and the existence of the two versions).

38. The difference was first noted and described in van Schewick (2004, Chapter 6); van Schewick (2010a, p. 59). The terminology - ‘narrow’ vs ‘broad’ - reflects the differences in scope between the two versions (van Schewick, 2004). For a comparison of the two versions and a discussion of the rationale for distin­guishing between the two, see van Schewick (2010a, pp. 58-9, 75-81, 377-9, 408 fn. 67). In the literature, most older texts referring to ‘the end-to-end arguments’ simply quote either the narrow or the broad version (see, e.g., the references cited by van Schewick, 2010a, p. 59). For two exceptions, see Moors (2002); Kempf and Austein (2004).

39. van Schewick (2010a, pp. 5, 58-9, 79-80) (discussing examples).

40. For a detailed analysis of the broad version, see van Schewick (2010a, pp. 67-75, 378-9). For an analy­sis of the narrow version and its relationship to the Internet’s original architecture, see van Schewick (2010a, pp. 60-67, 90-96, 103, 377-8, 380). On the two versions of the end-to-end arguments in the current Internet, see, for example, Saltzer (1999); Reed (2000); Blumenthal and Clark (2001); Clark and Blumenthal (2011); van Schewick (2010a, pp. 383-7, 388-9) (with references to the literature).

41. Blumenthal and Clark (2001, p. 71).

42. Reed et al. (1998, p. 69).

43. See, for example, Reed et al. (1998, pp. 69-70); Blumenthal and Clark (2001, pp. 71-2); Clark et al. (2005, p. 471); van Schewick (2010a, pp. 68-9).

44. Blumenthal and Clark (2001, p. 80) (citing personal communication with Jerome Saltzer).

45. See, for example, Reed et al. (1998, pp. 69-70); Reed (2000); Clark et al. (2005, p. 472); van Schewick (2010a, pp. 69-71).

46. Application autonomy is the idea that an application or higher layers close to it know best what services they need and should therefore be responsible for meeting these needs. On the broad version and applica­tion autonomy, see, for example, Reed et al. (1998, p. 70); van Schewick (2010a, pp. 71-2). On the broad version and reliability, see, for example, Blumenthal and Clark (2001, p. 71); Clark et al. (2005, p. 472); van Schewick (2010a, p. 72).

47. van Schewick (2010a, pp. 68-75, 355-71, 378-9) (discussing the trade-off underlying the broad version).

48. van Schewick (2010a, pp. 72-5, 217-18); van Schewick (2015a, pp. 128-30, Box 19 and fn. 455); van Schewick (2004). See also, for example, Lemley and Lessig (1999, para. 17); Cerf (2006, pp. 1-4, 7); Reed (2010). Like any application-blind network, the resulting network is also application-agnostic. See note 34.

49. See note 33 and accompanying text.

50. See van Schewick (2010a, pp. 88-90, 379). On the Internet layer as a portability layer or, in Clark’s ter­minology, ‘spanning layer’, see also Clark (1997, pp. 134 5); Computer Science and Telecommunications Board and National Research Council (2001, pp. 126-30). On relaxed layering with a portability layer as a design principle, see van Schewick (2010a, pp. 47-8).

51. This terminology follows the terminology of the TCP/IP reference model used by the Internet Engineering Task Force (IETF). See, for example, Comer (2000, pp. 183-5); Tanenbaum and Wetherall (2011, pp. 45-8). For an overview of the different layers of the Internet’s architecture and the main protocols at these layers, see van Schewick (2010a, pp. 83-8).

52. See note 29 and accompanying text.

53. See, for example, Computer Science and Telecommunications Board and National Research Council (2001, pp. 126-30); Peterson and Davie (2012, pp. 35-6).

54. See van Schewick (2010a, pp. 96-103, 110-12, 380-81). According to the networking literature, the end-to-end arguments are among the few design principles underlying the architecture of the Internet. See, for example, Carpenter (1996, section 2); Reed et al. (1998, p. 70); Computer Science and Telecommunications Board and National Research Council (2001, pp. 36-8); Braden et al. (2000, p. 15); Blumenthal and Clark (2001, p. 71); Clark et al. (2005, section VI.A). These texts effectively describe the broad version of the end-to-end arguments. See, for example, the citations in van Schewick (2010a, pp. 410-11 fn. 69). On the impact of the narrow version on the Internet’s original architecture, see van Schewick (2010a, pp. 90-96, 380). On the impact of the two versions of the end-to-end arguments on application design, see van Schewick (2010a, pp. 103, 107-10).

55. For an overview of the initial proposal, see van Schewick (2010a, pp. 94-5). For technical descriptions of the initial monolithic protocol, see Cerf and Kahn (1974) (the first published description); Cerf (1977) and Cerf and Postel (1978).

56. See, for example, van Schewick (2010a, p. 97)

57. van Schewick (2010a, pp. 97-8, 102-3).

58. van Schewick (2010a, pp. 97-8).

59. The decision to split the initial monolithic protocol was made at a meeting of the researchers involved in the design of the Internet Transmission Control Protocol on 30-31 January 1978 at the Information Sciences Institute in Marina Del Rey. See, for example, Postel (1978) (describing the results of the meeting). See also Abbate (1999, pp. 129-30); van Schewick (2010a, Box 3.3, pp. 99-100). For a detailed discussion of the considerations that motivated this decision, for a description by David Reed (one of the authors of the papers describing the end-to-end arguments) of the events leading to the decision, and for references to the literature underlying the arguments in the text, see van Schewick (2010a, pp. 96-101 and Box 3.3, pp. 99-100).

60. In addition, the choice to offer only unreliable datagram service at the Internet layer was motivated by the designation of the Internet layer as the portability layer and the desire to enable applications to survive partial network failures. See, for example, Clark (1988, pp. 107-10); van Schewick (2010a, pp. 101-2).

61. For a short description of the Internet Protocol, see van Schewick (2010a, pp. 85-6).

62. For short descriptions of TCP and UDP, see van Schewick (2010a, pp. 86-7).

63. See also note 48 and accompanying text. Like any application-blind network, the Internet’s original architecture was also application-agnostic. See note 34.

64. Again, I use the term applications as shorthand for applications, content, services, and uses.

65. The Internet’s architecture also influences the environment for innovation in other layers of the Internet’s architecture. See, for example, van Schewick (2010a, pp. 106-7, 151-4, 157-63, 366, 382-3) discussing the impact of Internet architecture on innovation in the Internet layer and pp. 88-90, 148-55 discussing the impact of Internet architecture on innovation in the link layer.

66. van Schewick (2010a, pp. 140-41).

67. On the term core of the network, see notes 23 to 24 and accompanying text.

68. I use the terms network providers to describe economic actors who provide Internet access or transport services, regardless of whether they are vertically integrated into the operation of the network infrastruc­ture or not. See also note 5.

69. van Schewick (2010a, pp. 204, 211, 293); Cerf (2006, pp. 1-4). See also Balkin (2009) (focusing on the social, cultural, and political implications of innovation without permission).

70. See van Schewick (2010a, pp. 138-48, 204-5, 289-90). See also Benkler (2000, pp. 565-8); Balkin (2009) (both focusing on the social, cultural, and political implications).

71. The overall costs of developing an application and making it available to others on an ongoing basis differ depending on the type of application. While these costs will be lower for applications that run entirely on users’ machines than for applications that need servers run by the developer or provider of the application, recent developments have drastically reduced the minimum level of investment needed to develop and operate server-based applications. See van Schewick (2010a, pp. 143-4).

72. van Schewick (2010a, pp. 204-13, 292-3). See generally van Schewick (2010a, pp. 115-18, 165-6) (discussing the relationship between architecture and who can innovate).

73. van Schewick (2010a, pp. 205, 206). This seems to be an important motivation of contributors to open-source projects. Lakhani and Wolf (2005, pp. 12-16). See also Benkler (2002).

74. van Schewick (2010a, pp. 205-6, 334-5).

75. The term deploying an application denotes all steps that a user and the operator of the network must perform before an application can be used for the first time. See van Schewick (2010a, pp. 137-8).

76. van Schewick (2010a, pp. 144, 152-5, 293-5, 362-4); Cerf (2006, pp. 1-3, 7). On the importance of user choice for the Internet’s social, cultural, and political potential, see, for example, Balkin (2009); van Schewick (2010a, pp. 359-65).

77. Not every application of the broad version automatically results in decentralized control by users (instead of centralized control by a few network providers); it only results in decentralized control if end users control the end hosts. Thus, control over a network can be centralized even if the network is designed according to the broad version, if a central entity controls the end hosts. For example, an enter­prise network based on the TCP/IP suite where applications can only be installed by system administra­tors is based on the broad version, but centrally controlled. See van Schewick (2010a, pp. 72, 414 fn. 118, 387). In the commercial Internet, users generally control the end hosts (ibid., pp. 152-5, 362-4). In the past, network providers and handset providers often controlled the handsets on mobile networks (ibid., pp. 390-91). On the impact of specialized appliances controlled by the manufacturer or the network provider on user choice and application innovation under an architecture based on the broad version, see ibid., p. 387; Gillett et al. (2001); Zittrain (2008).

78. van Schewick (2010a, pp. 217-18).

79. Discrimination not only encompasses differential handling of packets, but also other forms of conduct that make an application or class of application relatively more attractive to use than others, including, for example, application-specific pricing or differential counting of applications against the monthly usage caps. See van Schewick (2015a, pp. 30-33); van Schewick (2015c, pp. 1-3); van Schewick (2016).

80. van Schewick (2010a, pp. 293-294). The effect of application-awareness on control over application innovation, deployment, and use is explained in more detail below. See subsections 14.5.3.3 Control over Application Innovation and 14.5.3.4 Control over Application Deployment and Use.

81. See the analysis in subsections 14.5.1.1 Control over Application Innovation and 14.5.1.4 Control over Application Deployment and Use.

82. In a network that is application-blind, but contains application-specific functionality in the network’s core, the network may need to be changed before a new application can run on the network. Once the network is changed to enable an application, however, the network’s application blindness prevents the network provider from interfering with the applications on its network. In such a network, network providers control application innovation and deployment, but users control which applications are used (van Schewick, 2010a, pp. 293-4) (discussing a purely opaque network, i.e., a network that is application­blind, but not general).

83. van Schewick (2010a, pp. 270-73). See also Lessig (2008, pp. 7-8).

84. For two publicly documented examples, see van Schewick (2008b, p. 2) and the letter from the founders of the online video company Zediva to the FCC (Srinivasan and Gupta, 2010, pp. 1-2).

85. van Schewick (2010a, pp. 222-5). See also Whinston (1990, pp. 840, 850-52); Farrell and Katz (2000); Farrell and Weiser (2003, pp. 89, 100-105).

86. For a detailed analysis of incentives to block, see, for example, van Schewick (2010a, pp. 222-64, 275-8) (increase profits), pp. 266-270 (block unwanted content), pp. 264 6 (manage congestion); van Schewick (2008a, pp. 5-6) (manage congestion).

87. van Schewick (2010a, p. 225). For a more detailed analysis of the costs of exclusionary conduct see van Schewick (2010a, p. 259-4); van Schewick (2015a, pp. 83-99) (discussing the factors affecting customers’ ability to switch providers of Internet access service).

88. van Schewick (2010a, p. 273, 277); van Schewick (2015a, pp. 56-7) (discussing examples). The impact of blocking on application developers’ incentives to innovate stems from the blocking as such and is independent of whether the network provider participates in the market for the application or not. By contrast, US antitrust law only condemns discriminatory conduct in the market for a specific application if the network provider participates in that market or is affiliated with a participant in that market. See van Schewick (2015a, p. 56).

89. See van Schewick (2010a, pp. 251-5, 264 70); Frischmann and van Schewick (2007, pp. 412-16). This chapter focuses on the impact of discrimination on application developers’ incentives to innovate. To reduce application developers’ incentives to innovate, the exclusionary conduct does not need to drive them from the market; it suffices if it reduces their profits. By contrast, scholars who evaluate discrimina­tory conduct within a framework based on US antitrust law will only be concerned about discriminatory conduct if the conduct is reasonably capable of monopolizing the market for the affected application or the market for Internet services. For a detailed analysis of this difference and references to the literature, see van Schewick (2015a, pp. 58-60). See also Frischmann and van Schewick (2007, pp. 414 fn. 119, 416 fn. 128).

90. AT&T (2009, pp. 6-7); van Schewick (2010a, pp. 241-2); BEREC (2012, p. 8).

91. See, for example, US Department of Justice (2011, pp. 11, 14 20, 37-9).

92. Wu (2003, pp. 151-2, 165); van Schewick (2010a, p. 471 fn. 237).

93. Comcast Corporation (2008); RCN Corporation (2010); Schatz (2008); Dischinger et al. (2008) (all US); Parsons (2009) (Canada); Cellan-Jones (2009); Cooper (2013) (all UK); BEREC (2012); Kroes (2012) (all Europe).

94. van Schewick (2010a, pp. 267-70) (describing the examples with references to the literature).

95. See, for example, Litan and Singer (2007, pp. 552-4); Yoo (2007, pp. 504, 506, 511-15); Becker et al. (2010, p. 505); Cave et al. (2009, pp. 1-2).

96. The following two paragraphs are adopted from van Schewick (2015a, pp. 83-99). For a full discussion with detailed references to the literature, see ibid., pp. 60-61, 83-9. For an earlier discussion, see van Schewick (2010a, pp. 259-64).

97. Relative to markets in which Internet service providers do not face any competitors, competition in the market for Internet services may even increase Internet service providers’ incentives to block or discriminate. See generally van Schewick (2010a, pp. 255-9) and, regarding incentives to engage in discriminatory traffic management, Cooper (2013) (based on a case study of broadband traffic man­agement in the UK).

98. See, for example, Cooper (2013) (wireline Internet services in the UK); BEREC (2012) (European wire­line and mobile Internet services); Kroes (2012) (same); Parsons (2009) (wireline Internet services in Canada); van Schewick (2011b) (Verizon Wireless/tethering applications); van Schewick (2011a) (AT&T, Verizon Wireless, T-Mobile/Google Wallet); Ziegler (2012); Kang (2012) (AT&T/Apple Facetime). See also van Schewick (2015a, pp. 96-8) (summarizing the evidence). On the amount of competition in the market for Internet services in the USA and Europe, see van Schewick (2015a, pp. 88-9).

99. For the EU, see Articles 20 and 21 Directive 2002/22/EC of the European Parliament and of the Council of 7 March 2002, as amended by Directive 2009/136/EC of the European Parliament and of the Council of 25 November 2009 (Universal Service Directive). For the USA, see 47 C.F.R. §8.3. On the effect of disclosure rules on network providers’ incentives to discriminate, see van Schewick (2015a, pp. 83-99).

100. Any Internet service provider can charge fees to customers of its Internet access service, regardless of whether these customers are providers of applications or ‘normal’ end users. In the past, Internet users - both application providers and ‘normal’ end users - directly paid fees for Internet service only to their own Internet access provider.

101. A large body of literature discusses the different types of access fees and which, if any, regulatory inter­vention is needed. For a brief overview, see van Schewick (2010a, p. 4). For opponents of access fees, see, for example, Lee and Wu (2009); van Schewick (2010b); Economides and Tag (2012); Economides and Hermalin (2012); van Schewick (2015b, pp. 11-17). For proponents, see, for example, Hemphill (2008) and Schuett (2010, pp. 2-4); Faulhaber (2011, pp. 62-8) (both reviewing the economic literature on access fees from the perspective of a proponent of access fees).

102. The practice of an ISP not counting selected applications against a user’s monthly data cap is also called ‘zero rating’. On the social costs and benefits of the different kinds of zero rating (including a discussion of real-world examples), see van Schewick (2015c); van Schewick (2016).

103. van Schewick (2010a, pp. 278-80, 290-92; 2010b, pp. 2-3).

104. van Schewick (2010a, pp. 207-10, 211-13, 292-3; 2010b, pp. 1-6; 2015b, pp. 11-14).

105. Reducing or impeding the ability of innovators with little or no outside funding to develop new applica­tions may significantly reduce the amount and quality of application innovation. Throughout the history of the Internet, many important innovations (including eBay, Facebook, Yahoo!, Google, Apache Web Server, the World Wide Web, Flickr, and Blogger) have been developed by innovators of this type. See van Schewick (2010b, pp. 3-5); van Schewick (2010a, pp. 204-13, 310-14, 318-28, 334-45) (discussing the importance of different types of low-cost innovators, including many examples). As the comments by many start-ups in the FCC’s Open Internet Proceeding in 2014-15 show, low-cost innovators continue to be important sources of innovation today. See van Schewick (2015b, pp. 12-13 and Appendix ‘Internet Startups Need a Non-Discriminatory Internet’).

106. van Schewick (2010b, p. 3; 2010c, pp. 4 5, 6). On the consequences of allowing network providers to influence which application will be successful, see section 14.5.3.4 Control over Application Deployment and Use.

107. van Schewick (2010a, pp. 217-18, 273) (discussing the impact of architecture on available pricing strate­gies). For a detailed analysis of network providers’ incentives to engage in these strategies and of the impact on application developers and users, see van Schewick (2010a, pp. 273-5) (application-specific pricing), pp. 275-8 (price discrimination). Application-specific pricing may also be used to discriminate among applications or classes of applications (van Schewick, 2015a, pp. 32-3). For two real-world exam­ples of these strategies, see Allot Communications and Openet (2010, p. 7) (application-specific pricing) and Wu (2003, pp. 151-2, 165); van Schewick (2010a, p. 471 fn. 237) (price discrimination).

108. van Schewick (2010a, pp. 286-7). For a description of four prototypical examples, see van Schewick (2010a, pp. 287-9).

109. van Schewick (2010a, pp. 286, 382, 385-6).

110. van Schewick (2010a, pp. 74-5, 286-7, 371-2).

111. van Schewick (2010a, pp. 70, 264-6, 286, 372).

112. van Schewick (2010a, pp. 286, 382, 385-6).

113. van Schewick (2010a, pp. 286-7, 371-2). On deep packet inspection (DPI) in general, see Anderson (2007); Cooper (2011). For specific examples, see, for example, Cisco Systems (2005); Free Press (2010a); Allot Communications and Openet (2010); Talbot (2014). On the state of DPI deployment, see Free Press (2010b, pp. 141-51).

114. For an analysis of the impact of four prototypical architectures (end-to-end, partially controllable, fully controllable, core-centered) on the economic environment for application innovation, see van Schewick (2010a, pp. 287-95, 351-3).

115. van Schewick (2010a, pp. 115-18, 215-16).

116. See section 14.5.2 Application Blindness of the Network and van Schewick (2010a, Chapter 6).

117. van Schewick (2010a, pp. 138-40, 144 8). See the section 14.5.3.2 Size and Diversity of the Innovator Pool.

118. van Schewick (2010a, pp. 144-7, 289-91). By contrast, deviating from the broad version of the end- to-end arguments may increase network providers’ incentives to innovate at the application level. van Schewick (2010a, p. 291).

119. van Schewick (2010a, pp. 144-8, 152-5, 294-5, 352).

120. See generally van Schewick (2010a, pp. 115-18, 165-6).

121. See subsections 14.5.1.2 Cost of Application Innovation and 14.5.1.3 Size and Diversity of the Innovator Pool.

122. van Schewick (2010a, pp. 210-11, 292-3).

123. See notes 104 to 105 and accompanying text.

124. On the need for changes to the core of the network in an application-aware architecture and the impact on the costs of application innovation, see van Schewick (2010a, pp. 138-40, 144-8, 291).

125. van Schewick (2010a, pp. 210-13, 292-3).

126. For the full argument with references to the literature, see van Schewick (2010a, pp. 297-345).

127. van Schewick (2010a, pp. 298-301). Generally, see, for example, Nelson and Winter (1977, p. 47; 1982, pp. 389-90); North (1990, pp. 80-82); Merges and Nelson (1994, p. 6); Rosenberg (1994, pp. 87-108; 1996); Cohen and Malerba (2001). On the link between the end-to-end architecture of the Internet and evolutionary theories of innovation, see Wu and Lessig (2003, pp. 5-7); Wu (2003, pp. 145-6; 2004, section II.A).

128. van Schewick (2010a, pp. 297-345).

129. For example, Kirzner (1997); Shane (2000); Shane and Venkataraman (2000); Sarasvathy et al. (2003).

130. For example, Rosenberg (1996).

131. van Schewick (2010a, pp. 301-6).

132. For numerous examples from the history of the Internet, see van Schewick (2010a, pp. 301-10).

133. van Schewick (2010a, pp. 310-11).

134. van Schewick (2010a, pp. 300). The impact of differences among actors also increases with the complex­ity of the problem to be solved and with the heterogeneity of customers. Ibid.

135. van Schewick (2010a, pp. 116, 311-12).

136. van Schewick (2010a, pp. 312-14). For more on the differences among and specific value of different types of innovators with examples from the history of the Internet, see van Schewick (2010b, pp. 3-5); van Schewick (2010a, pp. 204-13, 310-14) (low-cost innovators), pp. 312-13, 319-28 (large companies), pp. 319-28 (new entrants), pp. 208-9, 312, 319, 328-34 (venture capitalists), pp. 206-7, 209, 319, 312-13 (self-funding or angel investors), pp. 205-6, 312-13, 319, 334 45 (users). See also note 105.

137. van Schewick (2010a, pp. 314-17).

138. van Schewick (2010a, pp. 318-19).

139. For a more nuanced analysis, see van Schewick (2010a, pp. 299-300, 349). See also Nelson and Winter (1982, pp. 389-90); Merges and Nelson (1994, p. 6); Rosenberg (1996, p. 353).

140. See section 14.5.1.1 Control over Application Innovation.

141. van Schewick (2010a, pp. 138-40, 293-4).

142. van Schewick (2010a, pp. 288-9, 293).

143. van Schewick (2010a, pp. 292-4, 345-6).

144. On the arguments in the text, see van Schewick (2010a, pp. 345-8) with references to the literature.

145. On this requirement, see note 77.

146. See section 14.5.1.4 Control over Application Deployment and Use.

147. van Schewick (2010a, pp. 152-5, 293-5, 362-4).

148. See van Schewick (2010a, p. 152-5, 294-5, 346, 349, 485 fn. 162).

149. van Schewick (2010a, pp. 299-300, 349).

150. In the terminology of evolutionary economics, different architectures create different selection environments.

151. van Schewick (2010a, pp. 350-51).

152. See van Schewick (2010a, p. 350-51) and section 14.5.2.1 Blocking or Discrimination.

153. See notes 89 to 99 and accompanying text.

154. van Schewick (2010a, p. 351).

155. See van Schewick (2010a, pp. 362-3); van Schewick (2008a, pp. 7-8). See also Cerf (2006, pp. 1-3, 7). On the importance of user choice for the Internet’s social, cultural, and political potential, see, for example, Balkin (2009); van Schewick (2010a, pp. 359-5).

156. Increasing the costs or reducing the expected benefits of application innovation will always reduce the amount of application innovation, regardless of the degree of uncertainty or user heterogeneity.

157. van Schewick (2010a, p. 351).

158. van Schewick (2010a, p. 356).

159. van Schewick (2010a, pp. 212-13). See also note 17.

160. van Schewick (2015a, pp. 124-31). Alternative non-discrimination rules may provide network providers with more or less flexibility regarding discrimination than the Internet’s original architecture. (In fact, the non-discrimination rule described in the text only requires the network to be application-agnostic; it does not require the network to be application-blind. See van Schewick, 2015a, pp. 24 5, Box 4, 130-31.) Thus, not all non-discrimination rules will necessarily recreate the economic environment created by an architecture based on the broad version. In addition, different non-discrimination rules will impose dif­ferent constraints on the evolution and operation of the network and will result in different costs of regu­lation. For an analysis of alternative proposals for non-discrimination rules, see van Schewick (2015a).

161. Application-specific criteria are criteria that depend on an application’s characteristics. Application­specific criteria include application (i.e., the specific instance of an application a user is using, e.g., Vonage vs Skype), application type (e.g., e-mail vs Internet telephony), the application-layer pro­tocol or transport-layer protocol the application is using (e.g., SIP vs Skype’s proprietary protocol, or TCP vs UDP), or the application’s technical requirements (e.g., latency-sensitive vs non-latency-sensitive applications). For a detailed description and analysis of the non-discrimination rule described in the text, see van Schewick (2015a, pp. 124 52).

REFERENCES

Abbate, J. (1999), Inventing the Internet, Cambridge, MA: MIT Press.

Allot Communications and Openet (2010), ‘Managing the unmanageable: Monetizing and controlling OTT applications. FierceLive! Webinar presentation’, attachment to the ex parte letter by Free Press In the Matter of Preserving the Open Internet, submitted 14 December, accessed 21 December 2015 at http://apps.fcc.gov/ ecfs/document/view?id=7020923750.

Anderson, N. (2007), ‘Deep packet inspection meets “net neutrality”, CALEA’, Ars Technica, 25 July, accessed 12 July 2015 at http://arstechnica.com/articles/culture/Deep-packet-inspection-meets-net-neutrality.ars.

Aoki, M. (2001), Towards a Comparative Institutional Analysis, Cambridge, MA: MIT Press.

AT&T (2009), AT&T response to Wireless Telecommunications Bureau letter, DA 09-1737 (31 July 2009), Letter to Federal Communications Commission, WT Dkt. No. RM-11361, August 21’, accessed 21 December 2015 at http://www.wired.com/images_blogs/business/2009/08/att-response-to-fcc.pdf.

Baldwin, C.Y. and K.B. Clark (2000), Design Rules: The Power of Modularity, Volume 1, Cambridge, MA: MIT Press.

Balkin, J.M. (2009), Testimony before the Federal Communications Commission at its Workshop on Speech, Democratic Engagement, and the Open Internet, Washington, DC: Federal Communications Commission.

Becker, G.S., D.W. Carlton and H.S. Sider (2010), ‘Net neutrality and consumer welfare’, Journal of Competition Law and Economics, 6 (3), 497-519. Beije, P. (1998), Technological Change in the Modern Economy. Basic Topics and New Developments, Cheltenham, UK and Lyme, NH, USA: Edward Elgar Publishing.

Benkler, Y. (2000), ‘From consumers to users: Shifting the deeper structures of regulation towards sustainable commons and user access’, Federal Communications Law Journal, 52 (3), 561-79.

Benkler, Y. (2002), ‘Coase’s penguin, or, Linux and the nature of the firm’, Yale Law Journal, 112 (3), 369-446.

Benkler, Y. (2006), The Wealth of Networks: How Social Production Transforms Markets and Freedom, New Haven, CT: Yale University Press.

BEREC(2012),‘BERECfindingsontrafficmanagementpracticesinEurope’,Riga:BodyofEuropeanRegulatorsfor ElectronicCommunications,BoR(12)30,accessed19July2015athttp://berec.europa.eu/eng/document_register/ subject_matter/berec/reports/45-berec-findings-on-traffic-management-practices-in-europe.

Blumenthal, M.S. and D.D. Clark (2001), ‘Rethinking the design of the Internet: The end-to-end arguments vs. the brave new world’, ACM Transactions on Internet Technology, 1 (1), 70-109.

Braden, R., D. Clark, S. Shenker and J. Wroclawski (2000), ‘Developing a next-generation internet architec­ture’, accessed 22 July at http://www.isi.edu/newarch/DOCUMENTS/WhitePaper.pdf.

Carpenter, B. (1996), ‘Architectural principles of the Internet’, Request for Comments 1958, Internet Engineering Task Force (IETF), accessed 22 July 2015 at http://www.rfc-editor.org/pdfrfc/rfc1958.txt.pdf.

Castilla, E.J., H. Hwang, E. Granovetter and M. Granovetter (2000), ‘Social networks in Silicon Valley’, in C.-M. Lee, W.F. Miller, M. Gong Hancock and H.S. Rowen (eds), The Silicon Valley Edge: A Habitat for Innovation and Entrepreneurship, Stanford, CA: Stanford University Press, pp. 218-47.

Cave, M., R. Collins and N. van Eijk et al. (2009), ‘Statement by European academics on the inappropriateness of imposing increased Internet regulation in the EU’, accessed 21 December 2015 at http://www.cerre.eu/ sites/cerre/files/Statement_by_European_academics_on_the_inappropriateness_of_imposing.pdf.

Cellan-Jones, R. (2009), ‘iPlayer: BBC v BT’, BBC News dot.life Blog, 2 June, accessed 22 July 2015 at http:// www.bbc.co.uk/blogs/technology/2009/06/iplayerbbc_v_bt.html.

Cerf, V.G. (1977), ‘Specification of Internet Transmission Control Program TCP (Version 2)’, IEN 5, accessed 22 July 2015 at http://www.rfc-editor.org/ien/ien5.pdf.

Cerf, V.G. (2006), ‘Testimony before the United States Senate, Committee on Commerce, Science, and Transportation, at its hearing on: network neutrality’, accessed 24 December 2015 at https://www.gpo.gov/ fdsys/pkg/CHRG-109shrg30115/html/CHRG-109shrg30115.htm.

Cerf, V.G. and R.E. Kahn (1974), ‘A protocol for packet network intercommunication’, IEEE Transactions on Communications, 22 (5), 637-48.

Cerf, V.G. and J.B. Postel (1978), ‘Specification of Internetwork Transmission Control Program, TCP Version 3’, IEN 21, Information Sciences Institute, University of Southern California.

Cisco Systems (2005), ‘Network-based application recognition and distributed network-based application rec­ognition feature guide (Cisco IOS Release 12.4(4) T)’, accessed 21 December 2015 at http://www.cisco.com/cZ en/us/td/docs/ios/12_2s/feature/guide/fsnbarad.html.

Clark, D.D. (1988), ‘The design philosophy of the DARPA Internet Protocols’, Computer Communication Review, 18 (4), 106-14.

Clark, D.D. (1997), ‘Interoperation, open interfaces, and protocol architecture’, in NII 2000 Steering Committee Computer Science and Telecommunications Board and Mathematics Commission on Physical Sciences, and Applications and National Research Council (eds), The Unpredictable Certainty: White Papers, Washington, DC: The National Academies, pp. 133-44.

Clark, D.D. and M.S. Blumenthal (2011), ‘The end-to-end argument and application design: The role of trust’, Federal Communications Commission Law Review, 63 (2), 357-90.

Clark, D.D., J. Wroclawski, K.R. Sollins and R. Braden (2005), ‘Tussle in cyberspace: Defining tomorrow’s Internet’, IEEE/ACM Transactions on Networking, 13 (3), 462-75.

Cohen, W.M. and F. Malerba (2001), ‘Is the tendency to variation a chief cause of progress?’, Industrial and Corporate Change, 10 (3), 587-608.

Comcast Corporation (2008), ‘Comcast Corporation description of current network management practices, Attachment A to Comcast Corporation’s Filing in the Matter of Formal Complaint of Free Press and Public Knowledge Against Comcast Corporation for Secretly Degrading Peer-to-Peer Applications submitted 19 September 2008, WC Dkt. No. 07-52’, accessed 21 December 2015 at https://downloads.comcast.net/docs/ Attachment_A_Current_Practices.pdf.

Comer, D.E. (2000), Internetworking with TCP/IP: Principles, Protocols, and Architectures, 4th edition, Upper Saddle River, NJ: Prentice Hall.

Computer Science and Telecommunications Board and National Research Council (2001), The Internet’s Coming of Age, Washington, DC: National Academy Press.

Cooper, A. (2011), ‘Doing the DPI dance’, in W Aspray and P. Doty (eds), Privacy in America: Interdisciplinary Perspectives, Lanham, MD: Scarecrow Press, Inc., pp. 139-65.

Cooper, A. (2013), ‘How competition drives discrimination: An analysis of broadband traffic management in the UK’, paper presented at the 41st Research Conference on Communication, Information and Internet Policy (TPRC’41), Arlington, VA.

DiMaggio, PJ. (1998), ‘The New Institutionalism: Avenues of collaboration’, Journal of Institutional and Theoretical Economics, 154 (4), 696-715.

DiMaggio, PJ. and W.W. Powell (1983), ‘The iron cage revisited: Institutional isomorphism and collective rationality in organizational fields’, American Sociological Review, 48 (2), 147-60.

Dischinger, M., A. Mislove, A. Haeberlen and K.P. Gummadi (2008), ‘Detecting BitTorrent blocking’, in Proceedings of the 8th ACM SIGCOMM Conference on Internet Measurement Conference (IMC^,08), pp. 3-8.

Dyer, J.H. (1996), ‘Specialized supplier networks as a source of competitive advantage: Evidence from the auto industry’, Strategic Management Journal, 17 (4), 271-91.

Dyer, J.H. (1997), ‘Effective interfirm collaboration: How firms minimize transaction costs and maximize transaction value’, Strategic Management Journal, 18 (7), 535-56.

Economides, N. and B.E. Hermalin (2012), ‘The economics of network neutrality’, RAND Journal of Economics, 43 (4), 602-29.

Economides, N. and J. Tag (2012), ‘Network neutrality on the Internet: A two-sided market analysis’, Information Economics and Policy, 24 (2), 91-104.

Farrell, J. and M.L. Katz. (2000), ‘Innovation, rent extraction, and integration in systems markets’, Journal of Industrial Economics, 48 (4), 413-32.

Farrell, J. and P.J. Weiser (2003), ‘Modularity, vertical integration, and open access policies: Towards a con­vergence of antitrust and regulation in the Internet Age’, Harvard Journal of Law and Technology, 17 (1), 85-134.

Faulhaber, G.R. (2011), ‘Economics of net neutrality: A review’, Communications and Convergence Review, 3 (1), 7-25.

Free Press (2010a), ‘Ex parte letter to the Federal Communications Commission, GN Dkt. No. 09-191’, 14 December.

Free Press (2010b), ‘Comments to the Federal Communications Commission, GN Dkt. No. 09-191’, 14 January.

Frischmann, B.M. and B. van Schewick (2007), ‘Network neutrality and the economics of an information superhighway: A reply to Professor Yoo’, Jurimetrics Journal, 47 (4), 383-428.

Furubotn, E.G. and R. Richter (2005), Institutions and Economic Theory: The Contribution of the New Institutional Economics, 2nd edition, Ann Arbor, MI: University of Michigan Press.

Garud, R. and A. Kumaraswamy (1993), ‘Changing competitive dynamics in network industries: An explora­tion of Sun Microsystems’ open systems strategy’, Strategic Management Journal, 14 (5), 351-69.

Gillett, S.E., W.H. Lehr, J.T. Wroclawski and D.D. Clark (2001), ‘Do appliances threaten Internet innovation?’, IEEE Communications Magazine, 39 (10), 46-51.

Granovetter, M. (1985), ‘Economic action and social structure: The problem of embeddedness’, American Journal of Sociology, 91 (3), 481-510.

Greif, A. (2006), Institutions and the Path to the Modern Economy: Lessons from Medieval Trade, Cambridge, UK: Cambridge University Press.

Hall, P. (1994), Innovation, Economics and Evolution. Theoretical Perspectives on Changing Technology in Economic Systems, New York: Harvester Wheatsheaf.

Hall, P.A. and R.C.R. Taylor (1996), ‘Political science and the three new institutionalisms’, Political Studies, 44 (5), 936-57.

Hemphill, C.S. (2008), ‘Network neutrality and the false promise of zero-price regulation’, Yale Journal on Regulation, 25 (2), 135-80.

Isenberg, D. (1997), ‘Rise of the stupid network: Why the intelligent network was once a good idea, but isn’t anymore. One telephone company nerd’s odd perspective on the changing value proposition’, Computer Telephony, August, 16, 18, 20, 24, 26.

Isenberg, D.S. (1998), ‘The dawn of the “stupid network”’, netWorker, 2 (1), 24-31.

Kang, C. (2012), ‘AT&T faces complaint over iPhone Facetime blocking’, Washington Post Technology Blog, 18 September, accessed 24 July 2015 at http://www.washingtonpost.com/blogs/post-tech/post/at&t-faces- complaint-over-iphone-facetime-blocking/2012/09/18/799c8650-0183-11e2-b257-e1c2b3548a4a_blog.html.

Kempf, J. and R. Austein (2004), ‘The rise of the middle and the future of end-to-end: Reflections on the evolu­tion of the Internet architecture’, Requestfor Comments3724, IETF, accessed 19 July 2015 at http://www.ietf. org/mail-archive/web/ietf-announce/current/msg00031.html.

Kirzner, I.M. (1997), ‘Entrepreneurial discovery and the competitive market process: An Austrian approach’, Journal of Economic Literature, 35 (1), 60-85.

Kroes, N. (2012), ‘Next steps on net neutrality - Making sure you get champagne service if that’s what you’re paying for’, 29 May, European Commission blog archive, accessed 24 July 2015 at http://ec.europa.eu/ archives/commission_2010-2014/kroes/en/blog/netneutrality.html.

Kurose, J.F. and K.W. Ross (2010), Computer Networking: A Top-Down Approach, 5th edition, Boston, MA: Pearson/Addison Wesley.

Lakhani, K.R. and R.G. Wolf (2005), ‘Why hackers do what they do: Understanding motivation and effort in free/open source software projects’, in J. Feller, B. Fitzgerald, S.A. Hissam and K.R. Lakhani (eds), Perspectives on Free and Open Source Software, Cambridge, MA: MIT Press, pp. 3-21.

Landes, W.M. and R.A. Posner (2003), The Economic Structure of Intellectual Property Law, Cambridge, MA: Belknap Press of Harvard University Press.

Langlois, R.N. and P.L. Robertson (1992), ‘Networks and innovation in a modular system: Lessons from the microcomputer and stereo component industries’, Research Policy, 21 (4), 297-313.

Lee, R.S. and T. Wu (2009), ‘Subsidizing creativity through network design: Zero-pricing and net neutrality’, The Journal of Economic Perspectives, 23 (3), 61-76.

Lemley, M.A. and L. Lessig (1999), ‘Ex Parte to Federal Communications Commission, CS Dkt. No. 99-251’, 10 November.

Lessig, L. (1998), ‘The New Chicago School’, Journal of Legal Studies, 27 (2), 661-91.

Lessig, L. (2001), The Future of Ideas: The Fate of the Commons in a Connected World, New York: Vintage Books.

Lessig, L. (2008), ‘Testimony before the United States Senate, Committee on Commerce, Science, and Transportation, at its Hearing on the Future of the Internet’, accessed 21 December 2015 at https://www.gpo. gov/fdsys/pkg/CHRG-110shrg74893/html/CHRG-110shrg74893.htm.

Litan, R.E. and H.J. Singer (2007), ‘Unintended consequences of net neutrality regulation’, Journal on Telecommunications and High Technology Law, 5 (3), 533-72.

Merges, R.P. and R.R. Nelson (1994), ‘On limiting or encouraging rivalry in technical progress: The effect of patent scope decisions’, Journal of Economic Behavior and Organization, 25 (1), 1-24.

Moors, T. (2002), ‘A critical review of “end-to-end arguments in system design”’, in Proceedings of the IEEE International Conference on Communications (ICCC 2002), pp. 1214-19.

Nelson, R.R. and S.G. Winter (1977), ‘In search of a useful theory of innovation’, Research Policy, 6 (1), 37-76.

Nelson, R.R. and S.G. Winter (1982), An Evolutionary Theory of Economic Change, Cambridge, MA: The Belknap Press of Harvard University Press.

North, D.C. (1990), Institutions, Institutional Change and Economic Performance, Cambridge, UK: Cambridge University Press.

Ostrom, E. (2005), ‘Doing institutional analysis: Digging deeper than markets and hierarchies’, in C. Menard and M.M. Shirley (eds), Handbook of New Institutional Economics, Dordrecht: Springer, pp. 819-48.

Parsons, C. (2009), ‘Summary of January 13, 2009 CRTC filings by major ISPs in response to inter­rogatory PN 2008-19 with February 9, 2009 updates’, accessed August 4, 2015 at http://www.christopher- parsons.com/PublicUpload/Summary_of_January_13_2009_ISP_filings_with_February_9_2009_Updates_ version_1.0(for_web).pdf.

Peters, B.G. (2005), Institutional Theory in Political Science. The New Institutionalism, 2nd edition, London: Continuum.

Peterson, L.L. and B.S. Davie (2012), Computer Networks: A Systems Approach, 5th edition, Burlington, MA: Morgan Kaufmann.

Postel, J. (1978), ‘Meeting Notes, 1 February 1978’, IEN 22.

Postel, J. (1981a), ‘Internet Protocol. DARPA Internet program protocol specification’, Request for Comments 791, IETF, accessed 24 July at http://www.rfc-editor.org/rfc/rfc791.txt.

Postel, J. (1981b), ‘Transmission Control Protocol. DARPA Internet program protocol specification’, Request for Comments 793, IETF, accessed 19 July 2015 at http://www.ietf.org/rfc/rfc0793.txt.

Powell, W.W. and PJ. DiMaggio (eds) (1991), The New Institutionalism in Organizational Analysis, Chicago, IL: University of Chicago Press.

Powell, W.W. and S. Grodal (2005), ‘Networks of innovators’, in J. Faberberg, D.C. Mowery and R.R. Nelson (eds), The Oxford Handbook of Innovation, Oxford, UK: Oxford University Press, pp. 56-85.

RCN Corporation (2010), ‘Ex parte letter to Federal Communications Commission, GN Dkt. No. 09-191’, 7 May, accessed 21 December 2015 at http://apps.fcc.gov/ecfs/document/view?id=7020916499.

Reed, D.P. (2000), ‘The end of the end-to-end argument’, accessed 19 July 2015 at http://www.reed.com/dpr/ locus/Papers/endofendtoend.html.

Reed, D.P. (2010), ‘A response to Barbara van Schewick: Code needs (only a little) help from the law’, dpr, 15 December, accessed 19 July 2015 at http://www.reed.com/blog-dpr/?p=85.

Reed, D.P., J.H. Saltzer and D.D. Clark (1998), ‘Commentaries on “active networking and end-to-end arguments”’, IEEE Network, 12 (3), 69-71.

Rosenberg, N. (1994), Exploring the Black Box. Technology, Economics and History, Cambridge, MA: Cambridge University Press.

Rosenberg, N. (1996), ‘Uncertainty and technological change’, in R. Landau, T. Taylor and G. Wright (eds), The Mosaic of Economic Growth, Stanford, CA: Stanford University Press, pp. 334 56.

Saltzer, J.H. (1999), ‘“Open access” is just the tip of the iceberg’, 22 October, accessed 19 July 2015 at http://web. mit.edu/Saltzer/www/publications/openaccess.html.

Saltzer, J.H., D.P. Reed and D.D. Clark (1981), ‘End-to-end arguments in system design’, 2nd International Conference on Distributed Computing Systems, 8-10 April, Paris, pp. 509-12.

Saltzer, J.H., D.P. Reed and D.D. Clark (1984), ‘End-to-end arguments in system design’, ACM Transactions on Computer Systems, 2 (4), 277-88.

Sanchez, R. (1995), ‘Strategic flexibility in product competition’, Strategic Management Journal, 16 (Special Issue, Summer), 135-59.

Sarasvathy, S.D., N. Dew, S.R. Velamuri and S. Venkataraman (2003), ‘Three views of entrepreneurial oppor­tunity’, in Z.J. Acs and D.B. Audretsch (eds), Handbook of Entrepreneurship Research, Boston, MA: Kluwer Academic Publishers, pp. 141-60.

Saxenian, A.L. (1994), Regional Advantage: Culture and Competition in Silicon Valley and Route 128, Cambridge, MA: Harvard University Press.

Schatz,A.(2008),‘Coxabouttofeelwrathofnetneutralityactivists’, WSJBlogs: Washington Wire,15 May,accessed 19 July 2015 at http://blogs.wsj.com/washwire/2008/05/15/cox-about-to-feel-wrath-of-net-neutrality-activists/.

Schuett, F. (2010), ‘Network neutrality: A survey of the economic literature’, Review of Network Economics, 9 (2), 1-15.

Scott, W.R. (2000), Institutions and Organizations, 2nd edition, Thousand Oaks, CA: Sage Publications.

Shane, S. (2000), ‘Prior knowledge and the discovery of entrepreneurial opportunities’, Organization Science, 11 (4), 448-69.

Shane, S. and S. Venkataraman (2000), ‘The promise of entrepreneurship as a field of research’, Academy of Management Review, 25 (1), 217-26.

Srinivasan, V. and V Gupta (2010), ‘Ex parte letter to Federal Communications Commission, GN Dkt. No. 09-191’, 10 December.

Sterbenz, J.P.G. and J.D. Touch (2001), High-Speed Networking. A Systematic Approach to High-Bandwidth Low-Latency Communication, New York: John Wiley and Sons, Inc.

Sunstein, C.R. (1996), ‘On the expressive function of law’, University of Pennsylvania Law Review, 144 (5), 2021-53.

Takeishi, A. (2001), ‘Bridging inter- and intra-firm boundaries: Management of supplier involvement in automobile product development’, Strategic Management Journal, 22 (5), 403-33.

Talbot, D. (2014), ‘Net neutrality quashed: New pricing schemes, throttling, and business models to follow’, MIT Technology Review, 14 January, accessed 19 July 2015 at http://www.technologyreview.com/ news/523606/net-neutrality-quashed-new-pricing-schemes-throttling-and-business-models-to-follow/.

Tanenbaum, A.S. and D.J. Wetherall (2011), Computer Networks, 5th edition, Boston, MA: Prentice Hall.

Thelen, K. (1999), ‘Historical institutionalism in comparative politics’, Annual Review of Political Science, 2 (1), 369-404.

Ulrich, K. (1995), ‘The role of product architecture in the manufacturing firm’, Research Policy, 24 (3), 419-40. US Department of Justice (2011), ‘Competitive Impact Statement to United States District Court for the District of Columbia, United States of America, State of California, State of Florida, State of Missouri, State of Texas, and State of Washington v. Comcast Corp., General Electric Co. and NBC Universal, Inc., 18 January’, accessed 21 December 2015 at http://www.justice.gov/atr/case-document/ competitive-impact-statement-72.

Uzzi, B. (1996), ‘The sources and consequences of embeddedness for the economic performance of organiza­tions: The network effect’, American Sociological Review, 61 (4), 674-98.

van Schewick, B. (2004), ‘Architecture and innovation: The role of the end-to-end arguments in the original Internet’, PhD dissertation, Technical University Berlin, Germany.

van Schewick, B. (2008a), ‘Official testimony at the Federal Communications Commission’s second en banc hearing on broadband management practices’, Stanford, CA: Stanford University, accessed 21 December 2015 at https://transition.fcc.gov/broadband_network_management/041708/vanschewick-written.pdf.

van Schewick, B. (2008b), ‘Oral testimony at the Federal Communications Commission’s second en banc hearing on broadband management practices’, Stanford, CA: Stanford University, accessed 21 December 2015 at http://cyberlaw.stanford.edu/publications/oral-testimony-federal-communications-commission%E2%80% 99s-second-public-en-banc-hearing.

van Schewick, B. (2010a), Internet Architecture and Innovation, Cambridge, MA: MIT Press.

van Schewick, B. (2010b), ‘Opening statement at the Federal Communications Commission’s Workshop on Approaches to Preserving the Open Internet’, Center for Internet and Society, accessed 21 December 2015 at http://cyberlaw.stanford.edu/publications/opening-statement-federal-communications-commission %E2%80%99s-workshop-approaches-preserving.

van Schewick, B. (2010c), ‘Opening statement at the Federal Communications Commission’s Workshop on Innovation, Investment and the Open Internet in Cambridge, MA, WC Dkt. No. 07-52, GN Dkt. No. 09-191’, accessed 6 January 2016 at http://cyberlaw.stanford.edu/publications/opening-statement-federal- communications-commission%E2%80%99s-workshop-innovation-investment.

van Schewick, B. (2011a), ‘Is Verizon Wireless illegally blocking Google Wallet? It’s time for the FCC to inves­tigate’, Internet Architecture and Innovation Blog, 19 December, accessed 19 July 2015 at https://netarchitec- ture.org/2011/12/is-verizon-wireless-illegally-blocking-google-wallet-its-time-for-the-fcc-to-investigate/.

van Schewick, B. (2011b), ‘Public interest requires public input: Verizon/Android tethering’, Internet Architecture and Innovation Blog, 30 June, accessed 19 July 2015 at https://netarchitecture.org/2011/06/ public-interest-requires-public-input-verizonandroid-tethering/.

van Schewick, B. (2015a), ‘Network neutrality and quality of service: What a non-discrimination rule should look like’, Stanford Law Review, 67 (1), 1-166.

van Schewick, B. (2015b), ‘The case for meaningful network neutrality rules, submitted as an attachment to an ex parte letter to the Federal Communications Commission, GN Dkt. No. 14-28’, 20 February.

van Schewick, B. (2015c), ‘Zero-rating and network neutrality, submitted as an attachment to an ex parte letter to the Federal Communications Commission, GN Dkt. No. 14-28’, 19 February.

van Schewick, B. (2016), ‘T-Mobile’s Binge On violates key net neutrality principles, submitted as an attach­ment to an ex parte letter to the Federal Communications Commission, GN Dkt. No. 14-28’, 29 January.

Whinston, M.D. (1990), ‘Tying, foreclosure, and exclusion’, American Economic Review, 80 (4), 837-59.

Wu, T. (2003), ‘Network neutrality and broadband discrimination’, Journal on Telecommunications and High Technology Law, 2 (1), 141-75.

Wu, T. (2004), ‘The broadband debate: A user’s guide’, Journal on Telecommunications and High Technology Law, 3 (1), 69-95.

Wu, T. and L. Lessig (2003), ‘Ex parte submission to the Federal Communications Commission, CS Dkt. No. 02-52’, 22 August.

Yoo, C. (2007), ‘What can antitrust contribute to the network neutrality debate?’, International Journal of Communication, 1, 493-530.

Ziegler, C. (2012), ‘AT&T only allowing Facetime over cellular on mobile share plans, no extra charge’, The Verge, 17 August, accessed 19 July 2015 at http://www.theverge.com/2012/8/17/3250228/ att-facetime-over-cellular-ios-6-mobile-share.

Zittrain, J. (2008), The Future of the Internet and How to Stop It, New Haven, CT: Yale University Press.

15.