The Internet was designed to survive a nuclear war. But can it withstand the perils created by its own rapid growth?

At first glance, it’s hard to imagine that the Internet—in its current form, at least—could possibly be endangered. Total users are increasing at 120 percent per year. World Wide Web hosts are growing at 100 percent per year. Total spending on Internet products and services is rising at 110 percent per year. And the critical mass of users finally in place is attracting businesses that are making real commercial transactions—and consumers who are joining fledgling electronic communities.

Behind the scenes, however, the huge growth in demand from users shifting to ever more complicated applications, combined with rapid innovation, incomplete standards, fragmented competition, and unrealistic pricing policies, is producing an environment of wrenching and unpredictable change—one with which existing network service architectures are less and less able to cope. As users begin to flee increasingly degraded Internet performance in search of more specialized services, the many networks that make up the Internet are drifting further and further apart.

This balkanization of the Internet into multiple interconnected network families, each with a distinct user community and differentiated capabilities, means that the flat pricing regime that has encouraged wasteful use and discouraged responsible network engineering will soon be gone. Instead, users will be charged for what they use. Gone, too, will be the painfully slow response times, the lost messages, and the network outages that have become all too common on today’s Internet. Users will get what they pay for—even if they have to pay a lot.

The coming breakup of the Internet will affect more than just users: there will be important implications for service providers, technology providers, content providers, and regulators as well. Companies that anticipate and leverage these changes can help shape the future to their advantage, while those that cling to the failing communal Internet model will quickly find themselves left behind.

What’s wrong?

Since 1969, the Internet has grown to more than 5 million host computers, serving more than 30 million users on six continents. During this growth, the mix of traffic on the Internet has shifted dramatically, from simple transfers of files between host computers to bandwidth-intensive web surfing. Real-time video and voice telephony are just around the corner. It is far from clear the communal Internet can continue to support such explosive growth.

Bob Metcalfe, the inventor of Ethernet and founder of 3Com, predicted that the Internet would collapse before the end of 1996.¹ While this dire prediction has not come true, there have been clear signs of danger. NETCOM, one of the largest Internet service providers (ISPs), went out of service for 13 hours, stranding more than 400,000 users. America Online and its 6.2 million users were down for 19 hours. Both of these outages were caused by human error, but they can also be attributed to the boggling complexity of the network and the lack of any central control or accountability.

Beyond its effect on network reliability, the rapid growth in traffic and complexity has also affected everyday performance. Bottlenecks occur regularly, not only at popular content servers, but also at the backbone routers that direct Internet traffic and at the network access points (NAPs) where ISPs exchange traffic with one another. Packets must queue for tens to hundreds of milliseconds at each router before being passed to the next one. Multiply this by the 20 to 60 routers necessary for a round trip and the result is multisecond delays.

It is not uncommon for heavily congested routers to drop more than 10 percent of the packets that pass through them

Packet loss magnifies this effect. Noel Chiapa, the inventor of the multiprotocol router, suggests that it is not uncommon for heavily congested routers to drop more than 10 percent of the packets that pass through them. After it realizes that they are missing, the receiving computer will ask for these packets to be retransmitted, adding an additional round trip to the response time and exacerbating network congestion. Worse yet, some of the lost packets may contain important changes to the routing tables used by the routers to determine the destination of each packet on the Internet. Thus, portions of the Internet may become "invisible" to other portions until the routers readjust. Network abuse is yet another problem. With the touch of a button, an e-mail mass marketer can send unwanted material to millions of Internet subscribers for free—a practice known as "spamming." Hardcore Internet users have responded with electronic attacks that may disable the spammer, but in reality often hit innocent parties.

These problems mean that the needs of many actual and potential users are not being met by today’s Internet. Response time is slow, availability is unreliable, security is unproven, content is hard to find. Content providers have the same complaints, and few have found a money-making formula.

Service vendors are hit too. Though heavy investment seems to be required, ISPs are losing money and have little incentive to invest. They are paid a flat rate for best-efforts access, not for quality of service. Why not just dump the traffic on someone else and blame that network for poor performance?

Finally, networking equipment vendors, the biggest beneficiaries of the Internet’s growth, face substantial uncertainty. New technology platforms may ease the problems plaguing the Internet, but technical standards have not been set, and the mechanism for establishing them is unclear.

Why haven’t these problems been fixed?

While the unexpected explosion in demand is the cause of the Internet’s current predicament, and technology and investment are undoubtedly the solution, the path out is strewn with hurdles. Three categories of obstacles stand as the legacy of the Internet’s birth as a government-sponsored tool for the scientific community. Specifically, the environment lacks market pricing feedback to reward investment; it lacks efficient mechanisms to evolve technical standards; and it remains a commodity in the face of customer needs for differentiated services.

Market pricing

Network engineers tell horror stories of US college students holding round-the-clock video conferences with friends on the other side of the continent

The poor performance of the Internet is driven in large part by the lack of pricing feedback. Paying a traditional flat fee for access, users have no incentive to conserve bandwidth. Network engineers tell horror stories of US college students holding round-the-clock video conferences with friends on the other side of the continent. Internet telephony will exacerbate the problem, tying up precious transoceanic bandwidth as users arbitrage the artificial spread between international tariffs for voice and data.

Local ISPs also lack any incentive to conserve bandwidth. Paying a flat fee for connection to a national service provider’s (NSP’s) backbone network, local ISPs dump as much traffic onto the NSP as possible, particularly that destined for distant locations. Worse yet, some ISPs have been anonymously dumping traffic on NSP networks at NAPs without any agreement at all.

One by-product of flat-fee pricing is a lack of incentive for ISPs and NSPs to relieve congestion by adding capacity, as neither is paid for providing good service. Irresponsible providers are free riders, underinvesting, dumping traffic, and then blaming others for poor Internet performance. Meanwhile, none of the largest NSPs (such companies as Sprint, MCI, and UUNet) currently makes profits on its Internet backbone services—a further disincentive to investment. Bandwidth is expensive, and unless NSPs have a mechanism for charging for it on a usage- and distance-sensitive basis, it is unlikely to catch up with demand.

Flat pricing schemes may also hurt local telephone companies providing free local calling, used by Internet subscribers for dial-up access. Telephone networks are designed and priced to serve voice conversations, where typical calls last three to five minutes. Internet calls average 30 minutes and often stretch over 10 hours. According to telephone companies, growing Internet usage is beginning to tax local switches, and no mechanism exists to recover the costs of adding needed capacity. These companies claim that service in high-usage areas like Silicon Valley has suffered noticeably.

Standards mechanisms

Until 1995, Internet standards were coordinated by the National Science Foundation (NSF). In the commercial era, there is no single authority to define and enforce changes. Formal standards bodies, such as the Internet Engineering Task Force (IETF) and the ATM Forum, are ineffective in driving change to meet rapidly evolving user demand. In a way, standards bodies are like the United Nations: decisions take too long, specifications are vague and watered down, and industry participants are not obliged to follow suit.

In the Internet world, technological innovation occurs daily; the IETF, by contrast, meets quarterly. Innovation may consist of simple extensions to existing platforms, or entirely new technical approaches. Already, ISPs and NSPs are individually evaluating alternative technologies for providing service quality guarantees and for combining switching and routing in a single platform. Technology and service providers need some mechanism to coordinate the development and implementation of these innovations. The IETF and ATM Forum are more likely to act as standards acknowledging bodies than standards setting bodies.

Commoditization

The commercial Internet was born as a fragmented and undifferentiated service market

The commercial Internet was born as a fragmented and undifferentiated service market. This has commoditized Internet services and set multiple hurdles in the way of resolving problems. There is no clear leader to drive the resolution of pricing and engineering, and consensus is nearly impossible to reach. Users in any attractive segment are scattered across hundreds of networks, and cannot easily be aggregated. Nearly every Internet request has to traverse multiple networks, making it difficult to assess responsibility for weak performance and easy for poorly engineered networks to point fingers.

There are several reasons for this state of affairs. First, since the decommissioning of the NSFNET public backbone (see the boxed insert "Four myths about the Internet"), there have been no established metrics with which to compare ISP or NSP performance. Though the IETF has established a working group to try to define benchmarking criteria, this group has not yet been able to decide what metrics might be appropriate, much less devise a methodology for collecting and reporting them.

Second, Cisco Systems has about an 80 percent share of the installed base of routers in the networks that make up the Internet. Given this near monopoly in the most critical area of network technology, service vendors have found it hard to differentiate the value-added capabilities of their networks. Third, it is easy for a new ISP to enter the market by buying a flat-rate connection to an NSP and reselling it to multiple customers. Fourth, many large technology companies see the Internet as a strategic part of their long-term business, and are willing to pay to stay in the game. Finally, growth masks a multitude of sins, allowing players to stay in the market that might otherwise be forced out by competition.

What has to happen?

There is a way out. The Internet will change dramatically over the next year or two. What emerges will be a better, more robust Internet with market-based pricing, comprising a smaller number of interconnected but differentiated networks that adhere to standards determined by market competition. Gone will be $19.99 per month flat-rate Internet access; gone will be many of the smaller ISPs that do not offer added value to customers; gone will be backbone providers that do not meet clearly defined performance standards. This also means that engineers’ dreams of the integrated, do-everything network are likely to remain just dreams.

How will market pricing emerge?

NSPs have threatened to institute usage-sensitive settlements with those networks that have peering or transit relationships with them. Existing technology makes it impractical to meter and bill for usage or distance, but this is changing.

Corporate users are demanding improved reliability and performance, and are willing to pay for it. NSPs are already experimenting with offering multiple classes of service with guaranteed performance, using a new routing protocol called RSVP and special software. This technology has the added benefit of enabling traffic metering; NSPs will know who is sending what packets where, and can charge accordingly.

Similarly, local telephone companies are pushing to change the way they provide dial-up access to ISPs—and the way they charge for it. Internet users are being subsidized by ordinary telephone services, a state of affairs to which regulators will be sensitive, given the demographic profile of a typical Internet user. The combination of metering by NSP backbones and increased access costs is likely to drive ISPs to higher prices and user-sensitive rates.

Why will ISPs and NSPs consolidate?

Market-based pricing will drive out some players. Others will be acquired as local telephone companies enter the market. Unable to differentiate, yet others will exit the market. Incentives to aggregate addresses into large blocks will favor bigger networks. Economies of scale and scope in customer service, content hosting, network operations, and other functions will drive national players to acquire regional ones. All these factors will push the Internet service provider market to greater concentration.

How will providers differentiate?

Different customer segments have different needs. Already, MCI is building a parallel Internet with premium performance for corporate virtual private networks, while the scientific community is clamoring for its own Internet to meet its particular needs. @Home is building another parallel Internet with special features designed to take advantage of cable TV networks to deliver video-intensive high-speed consumer services. While all these networks will be an interconnected part of the Internet environment, they will be differentiated, offering special services available only to direct customers.

Differentiation will be achieved in several ways. First, special software overlays will be used to extend the capability of existing networks. An ISP serving the financial community may offer software-based security capabilities, for example. Second, networks may limit access to their physical infrastructure so as to control performance better for paying customers. Finally, networks may deploy special hardware adjuncts to deliver added-value services; for instance, new-generation switches or routers that make it possible to offer class-of-service performance guarantees.

How will standards be set?

Standards will be set by the market, with participants voting with their investment dollars and users with their feet

Standards will be set by the market, with participants voting with their investment dollars and users with their feet. To help control the risk of this uncertain environment, companies will begin to form loose affiliations with each other to support particular technical approaches. Some of these business webs will gain momentum; others will die out. Standards will emerge as winning business webs (like Wintel in the PC world) crowd out losers.

Why not a single integrated network?

Users do not care whether networks are integrated. They will not accept reduced performance for any of their network applications simply in order to achieve integration. Asynchronous transfer model (ATM), the solution most often touted for network integration, is ill suited to packet-based data networks like the Internet. Used in conjunction with TCP/IP, the overwhelming protocol choice of data networks, ATM adds a 20 to 30 percent overhead and increases the likelihood of delay from lost packets.

Voice and video traffic on packet-oriented data networks is similarly troublesome. Networks must be engineered with substantial excess bandwidth if performance is to approach that of traditional circuit-switched networks. Finally, the economics of migrating to a single network do not appear to be attractive, much less compelling.

What are the implications?

The evolution to an Internet composed of interconnected but specialized network families has important implications for service providers, technology providers, content providers, and regulators.

Service providers

Most smaller ISPs will not survive. Those that do will deliver superior service to clearly defined customer segments. They will affiliate with larger NSPs that can provide connectivity to content and access to other communities of interest on a national or global basis. Lacking both customer focus and scale, mid-sized regional ISPs will be particularly vulnerable. Larger ISPs seeking aggressively to expand their customer base, such as local telephone companies, may find them attractive candidates for acquisition.

Successful NSPs will tailor their networks to meet the specific needs of high-value customer segments, as with @Home’s construction of a new backbone designed to deliver consumer services over cable TV networks. This network architecture uses the high-speed access provided by cable modems in combination with content replicated and distributed to host computers nationwide to enable video applications not feasible with other Internet backbones. Similarly, NSPs BBN and PSINET are focusing on delivering guaranteed quality of service for large national business customers.

Content hosting will become an increasingly value-added service. The closer content servers are to customers, the better their performance. Moreover, more tightly defined customer segments will demand more tightly defined content environments. Content must be organized and edited to be valuable.

Technology providers

Hardware providers have been the early winners on the Internet. Cisco has the third-highest market capitalization on the NASDAQ exchange, behind only Microsoft and Intel. Similarly, small companies like Cascade and Ascend Communications have become large and profitable in months on the strength of their Internet-oriented products. The next three years may see such companies do even better, as market pricing gives NSPs the incentive to invest in growing and differentiating their networks.

The formation of business webs will lend momentum to new solutions to the needs of specic customer segments

Service differentiation may also break Cisco’s hold on the router market. The formation of business webs will lend momentum to new solutions to the needs of specific customer segments. New entrants will enjoy crucial opportunities to shape future standards and capture value if they are able to gain a position in a winning web.

Two different types of software are set to create and capture value in the heterogeneous but interlinked Internet. First, middleware software standards will be critical in defining the rules of inter-network connectivity. Users will venture outside their home networks on occasion, and there must be software to enable these communications. Second, differentiated networks offer an opportunity for specialized applications to exploit the particular capabilities of each network. One-size-fits-all approaches may miss the mark.

Content providers

The evolution of the Internet will be good news for content providers, making it easier to provide better service to more targeted user communities. User information will be more complete and more readily available. New network capabilities and better development tools will foster more compelling content. A robust hosting market will reduce investment in equipment and skills.

The biggest winners will be those providers that are able to create and manage electronic communities of interest.² The differentiated Internet will be far more hospitable to electronic communities, since customer segments will naturally gravitate to the physical networks targeted at their needs. As their loyalty grows, users will look to these communities for communication, information, and transactions. The providers that organize these electronic communities will capture more than a fair share of a rapidly expanding pie. Once established, it will be difficult for an electronic community to be displaced.

Regulators

Governments will play a critical role in the evolution of the Internet. Market pricing, consolidation, and differentiated networks can only emerge if the United States’ Federal Communications Commission, state public utilities commissions, and worldwide regulatory bodies allow them to happen. Efforts to mandate flat pricing, limit consolidation, or dictate standards will mean that the problems that plague the Internet will deepen.

An interventionist regulatory scenario might just make Bob Metcalfe’s prediction come true. The collapse of the Internet would set back electronic communications by many years. Perhaps a better Internet would rise in its place, but momentum would be lost. Users would be skeptical, service providers fearful, and content providers wary of getting burned a second time.

A market-priced, consolidated but differentiated Internet is the most likely direction of evolution, but it is not the only plausible path. Government policy could have a dramatic effect on the future. Similarly, the nature of technical innovation and competitive conduct is far from certain.

Few companies predicted the astounding growth of the Internet over the past three years. Those that prospered were prepared for change and flexible enough to adapt to the market as it evolved. The same will hold true for the next three years.