The changing patterns of Internet usage.

• Author: Yoo, Christopher S.

1. INTRODUCTION II. INTERNET PROTOCOL VIDEO A. Bandwidth and Quality of Service B. Congestion Management C. Multicasting D. Regulatory Classifications III. WIRELESS BROADBAND A. Bandwidth Limits and Local Congestion B. The Physics of Wave Propagation C. Congestion Management D. The Heterogeneity of Devices E. Routing and Addressing IV. CLOUD COMPUTING A. End-User Connectivity B. Data Center Connectivity C. Privacy and Security V. PROGRAMMABLE NETWORKING VI. PERVASIVE COMPUTING AND SENSOR NETWORKS VII. CONCLUSION I. INTRODUCTION

   The Interact unquestionably represents one of the most important technological developments in recent history. It has revolutionized the way people communicate with one another and obtain information and has created an unimaginable variety of commercial and leisure activities. Many policy advocates believe that the Internet's past success depended in no small part on its architecture and have argued that its continued success depends on preserving that architecture in the future. (1)

   Interestingly, many members of the engineering community see the Internet in starkly different terms. They note that the Internet's origins as a military network caused it to reflect tradeoffs that would have been made quite differently had the Internet been designed as a commercial network from the outset. (2) Moreover, engineers often observe that the current network is ill-suited to handle the demands that end users are placing on it. (3) Indeed, engineering researchers often describe the network as ossified and impervious to significant architectural change. (4) As a result, the U.S. government has launched a series of initiatives to support research into alternative network architectures. (5) The European Commission has followed a similar course, (6) and university-based researchers in both the United States and Europe are pursuing a variety of "clean slate" projects studying how the Internet might be different if it were designed from scratch today]

   This Essay explores some emerging trends that are transforming the way end users are using the Internet and examines their implications for both network architecture and public policy. Identifying future trends is inherently speculative and, in retrospect, will doubtlessly turn out to be mistaken in a number of important respects. Still, I hope that these ruminations and projections will yield some insights into the range of possible evolutionary paths that the future Internet may take.

2. INTERNET PROTOCOL VIDEO

   The development that has generated the most attention from policymakers and the technical community is the use of Internet-based technologies to distribute video programming. Over-the-top services (such as YouTube and Hulu) rely on the public Internet to distribute video. Other services, such as AT&T's U-verse, also employ the protocols developed for the Internet to distribute video, but do so through proprietary networks. Verizon's fiber-based FiOS service and many cable television providers already rely on these protocols to provide video on demand and are making preparations to begin using Internet-based technologies to distribute their regular video channels as well. Because these services are often carried in whole or in part by private networks instead of the public Internet, they are called Internet Protocol (IP) Video or IPTV. Industry observers have long predicted that video will represent an increasing proportion of total network traffic.

   The growing use of IP-based protocols to distribute video has raised a number of technical and policy challenges. Not only will the growth of IPTV require more bandwidth, it may also require more basic changes in the architecture and regulatory regimes governing the network.

   1. Bandwidth and Quality of Service

      Industry observers have long disputed how large the video-induced spike in network demand will actually be. A recent industry report estimates that Internet video now represents over one-third of all consumer Internet traffic and will grow to more than ninety percent of all consumer traffic by 2014. (8) Experts disagree about what the future holds. Some industry observers have long predicted the coming of a video-induced "exaflood" that would require a sharp increase in capital spending. (9) The Minnesota Internet Traffic Studies (MINTS) disagrees, pointing to the lack of any sign of such an upsurge in traffic. (10) Other observers challenge MINTS's conclusions, arguing that, in focusing solely on traffic patterns at public peering points, MINTS fails to take into account the sizable proportion of the overall traffic that now bypasses the public backbone. (11) Moreover, even if the shift to IP-based video distribution has not yet by itself created a spike in the demand for bandwidth, the wide-scale deployment of high-definition (and the looming emergence of ultra-high-definition), 3D, and multiscreen technologies may cause the rate of traffic growth to increase in the future.

      Aside from increased bandwidth, video requires network services that are qualitatively different in many ways from those required by the applications that formed the bulk of first-generation Internet usage. On the one hand, video is more tolerant of packet loss than web browsing and email. On the other hand, unlike the performance of email and web browsing, which depends solely on when the last packet is delivered, video quality depends on the timing with which every intermediate packet is delivered.

      Specifically, video is more sensitive to jitter, which is variations in spacing between intermediate packets in the same stream and which typically arises when a stream of packets traverses routers that are congested. Jitter can cause video playback to freeze temporarily, which degrades the quality of the viewers' experience.

      The usual solution to jitter is to delay playback of the video until the receiver can buffer enough packets to ensure that playback proceeds smoothly. This solution has the drawback of exacerbating another dimension of quality of service that is relevant for video, which is delay or latency, defined as the amount of time that it takes for playback to commence after it has been requested. Interestingly, viewers' tolerance of latency varies with the type of content being transmitted. While viewers of static video typically do not mind waiting five to ten seconds for playback to begin, such delays are not acceptable for interactive video applications, such as video conferencing. (12) Some content providers reduce latency by using data centers or content delivery networks to position their content in multiple locations, thereby shortening the distance between the content and end users. Storing content in multiple locations only works for static content that does not change. (13) It cannot work for interactive content, such as videoconferencing or online gaming, which changes dynamically.

      For interactive applications, the engineering community has focused on two other means for providing higher levels of quality of service. One solution is for network owners to overprovision bandwidth and switching capacity. When combined with distributed architectures for content delivery (such as caching and content delivery networks), this surplus capacity can give networks the headroom they need to handle any transient bursts in traffic without any congestion-related delays. (14) Overprovisioning is subject to a number of limitations, however. Wireless networks cannot simply add capacity to meet demand. Moreover, even networks that can increase bandwidth cannot do so instantaneously. Forecasting errors are inevitable, and in those instances where a network provider has failed to anticipate a key demographic shift or the emergence of a key application, device, or other complementary technology, it may sometimes find itself unable to expand capacity quickly enough to meet this increase in demand. (15) Over-provisioning also only increases the probability that particular traffic will pass through the network without delay. It does not guarantee the quality of service that any particular traffic will receive. (16) Finally, over-provisioning inherently requires networks to guarantee quality of service through capital expenditures (CapEx) rather than through operating expenditures (OpEx). As the difficulty in raising capital in the current economic downturn eloquently demonstrates, the relative cost of CapEx and OpEx solutions typically vary across time. Simple economics thus militate against locking network providers into one or the other option. (17)

      The other alternative to provide higher quality video service is to engage in increasingly sophisticated forms of network management that either reduce congestion or provide some means for providing higher levels of quality of service. Over the past two decades, the engineering community has developed a wide range of potential solutions, including Integrated Services (IntServ), (18) Differentiated Services (DiffServ), (19) MultiProtocol Label Switching (MPLS), (20) and Explicit Congestion Notification (ECN). (21) Other new initiatives, such as Low Extra Delay Background Transport (LEDBAT), also show promise. (22) Other engineers disagree with this approach, complaining that adding quality of service to the network would require devoting processing power in routers that would make the network too expensive and too slow. (23)

      Leading engineering textbooks recognize that the engineering community is split over which solution--overprovisioning or network management--would be better. (24) The fact that the engineering community has yet to reach consensus counsels against regulatory intervention mandating either approach.

   2. Congestion Management

      The advent of IPTV may also require fundamental changes to the way the network deals with congestion. The current approach to congestion management was developed in the late 1980s, shortly after the Internet underwent a series of...