2002: Creating the New Public Network

White paper proposing the creation of the Institute for the Promotion of the Internet Protocol Utility.

Transcript

1. Creating the New Public Network Tom Lyon pugs@ieee.org http://www.ipipu.org 8/30/2002 08:57:00 AM "Those who cannot remember the past are condemned to repeat it.” – George Santayana “Even if we do remember the past, we have to repeat it. But maybe we can fast forward to the good parts.” – me Overview The biggest problem facing the Internet today will not be solved by broadband, wireless, open access cable, a third wire, content control, or better security. The fundamental nature of the Internet presumes and provides universal connectivity, yet there has not been and will not be any business model to sustain this connectivity in the current competitive and regulatory frameworks. We’ll show that this is not at all unusual as an evolutionary phase of network utilities, and suggest what can be done to nurture the Internet through this period. In particular, we propose treating the IP level of the Internet as a publicly supported & controlled utility, so that competitive forces do not break the fundamental premise of connectivity. Puzzling Evidence… Computerworld, 12/13/2001: Planning on supporting full-time teleworkers or casual telecommuters with a secure virtual private network (VPN) over cable broadband service? Think again. Two of the major cable companies, Comcast Corp. and Cox Communications Inc., have boilerplate language buried in their residential service agreements that expressly forbids the use of a VPN over a residential broadband cable hookup. Network Magazine, 11/5/2001: In June 2001, when Tier 1 ISP PSINet (www.psinet.com) declared bankruptcy, it began to renege on peering commitments with 14 ISPs, including C&W. In retaliation, C&W killed its peering agreement with PSINet until the latter promised to bring exchanged traffic levels back to agreed upon levels. There are reports of users who couldn't reach some destinations on the Internet during the lapse, exactly the kind of problem that customers dread. And given the downturn hitting the industry, especially ISPs, right now, it's likely there'll be more incidents, which will probably force Tier 1s to enforce peering agreements more strictly. - 1 -
2. ITworld, 6/7/2002: KPNQwest, once a stock market darling worth US$40 billion, went bankrupt last Friday after a plan to sell some assets to meet its most urgent financial obligations failed. The company's 18 country, 25,000 kilometer fiber-optic network is said to be Europe's largest, carrying about 25 percent of the continent's Internet traffic. Center for Digital Democracy, 10/9/2001: Lucent says that one of the "first principles" shaping its "Mobile Internet platform" is the "flexibility to support diverse business models," including a walled garden approach. According to Lucent, its Mobile Internet Gateway enables a walled garden architecture because it "lets you add services to keep your customers close…." The product also provides for an open environment as well. But the Lucent Gateway allows the network operator to be the essential gatekeeper in determining how other applications are delivered through the system. Do we want our ISPs to determine what we can and can’t do with our Internet connection? Will the ISPs still be in business tomorrow? What is an Internet service anyway? And whose business is Internet transport? Network Utilities For over 100 years, economists have recognized the notion of network industries, where in the “core” of the network a natural monopoly seems to exist [Farrer]. These networks occur most commonly in transportation, utilities, and communications and result in situations where the networks are controlled for the “public good” either by public ownership or government regulation. These networks have very high “sunk” costs to achieve broad connectivity, but very low marginal costs, e.g., water, power, post, etc. The evolution of these utility style networks always starts at the local level, i.e., user A wishes to connect to user B to achieve better communication or less risk of shortages or any other pooling of interests. Early users are rarely concerned about business models, competitive vs. monopoly providers, etc., because they are happy to get some network as opposed to no network. As the benefits of the network become apparent, free enterprise jumps in to aid and benefit from the economic growth of the new network. However, as markets begin to be saturated, there emerge fundamental tensions between the advantages of universal connectivity versus the requirements of competition between providers. In the early 1900s, public dissatisfaction created the first regulation of both the electrical and the telephone networks. States began to mandate interconnectivity between competing telephone companies, which had used lack of interconnection as a tool to thwart competitors. Obviously, this led to unhappy customers who as a result sometimes had to have telephones from multiple companies. Due to the economies of scale in power generation in the electrical industry, states began to encourage consolidation of competitors into regional monopolies and began basic price regulation to control them. - 2 -
3. A very important aspect of these public utility networks is that the public network usually ends where private property begins. Thus, one pays for the water pipes & electrical wiring in one’s own home; large corporate or university campuses may have their own complex networks which need interconnection to the public networks. Thus public networks can be viewed as “Catenets,” networks that connect only to other networks, whether public or private. This model seems to become more true as networks evolve, e.g., the electrical utilities started as “lamp” companies. However, whether the network is public or private, the basic service delivered remains the same. These services are simple, commodity, lowest-common-denominator types of products, e.g., water, electric power, voice calls. Usually pricing is very simple also to reflect the simple nature of the service. Customers are known to prefer fixed rate pricing, but usage based is also common. What utilities don’t do is to price by ultimate use/application. The water you drink costs no more from the tap than the water you bathe in (though utilities may wish they were in the lucrative bottled water business). Electric power costs the same whether you use it for basic purposes like lighting, or for “luxury” purposes like powering your 800W audio system. Competitive network providers will rationally resist interconnection with smaller competitive networks because the interconnection is usually more beneficial to the smaller provider than the larger. Similarly, network providers discourage the notion that users can transition their networks from private services to public services; providers have been known to create both technical and legal obstacles to doing so. Smarter regulators will see that this is not in the public interest; in California, for instance, electrical providers must allow for “co-generation” of power from their users. (Although, as expected, P.G.&E. resists.) [Adelman] For the past 20 or 30 years, the trend in public utilities has been away from regulation and towards competition. Does this contradict the previous arguments? I think not – these utilities evolved under many years of regulation to be inefficient in both technology and management. I believe that the phase of maturity of the public networks along with their cost, stability, and efficiency are what will determine public satisfaction and thus regulatory climate. In theory, pure competition will lead to the most efficient networks; However, stability is often the victim of the competitive process. If users value stability more than efficiency, they will prefer a non-competitive solution. If there comes a point where inefficiencies are glaringly obvious, they may wish to introduce competition. In summary, competition in core networks can result in excess entry – too many providers such that none can make a return, and in insufficient scale – not enough profits to grow the network to new users. However, regulation can lead to inefficiencies, and monopoly control can lead to exploitation of users. The Old Public Network For most of the 20th century, the telephone networks of the U.S. and other countries enjoyed tremendous growth and enabled enormous economic gains as highly regulated, regional monopolies. Interconnection of these networks came about as the result of - 3 -
4. government actions, as international treaties, or state laws, leading to major on-going governmental or pseudo-governmental coordination efforts such as the I.T.U. The typical regulatory contract with the telephone operator granted the regional monopoly in return for universal service, common carriage, and no bundled services. Universal service required some level of effort on the operators part to reach every user within the region; common carriage required the operator to allow any kind of users or calls at published tariffs, in exchange for having no liability for those calls; no bundled services kept the large operators from using their customer relationship to sell services – this allowed new, innovative services to evolve more freely. The breakup of the Bell System in 1984 was the result of many years of studies and events going as far back as the 1966 FCC “Computer Inquiry”.[Cantelon] Even then, it was recognized that computer & data communication was a different beast than telephony, and the FCC required that it not be part of the regulated network in order to encourage innovation and to avoid dominance by the Bell companies. Since 1984, competition in the long distance market has resulted in much cheaper voice calls, and the long haul carriers have all invested heavily to meet and stimulate the growth in data traffic. Of course, fantastic new technologies like optical networks have played a huge part in the changes as well. So it is quite clear, that since 1984, there have been enormous efficiency gains in the telecom network. But since the telecom “bubble” of 1998-2000, we seem to have entered the excess entry phase of competition – massive overbuilds, price undercutting, etc., resulting in the financial destruction of many carriers. The 1996 Telecom act mandated open access at the local loop for competitors of the incumbent local exchange carriers (ILECs – like Pacific Bell); which attracted quite a bit of investment especially in the DSL “broadband” area. However, this deregulation model has failed – it results in complex business relationships, still too much power in the ILECs and too little oversight by the regulators. Meanwhile, the ILECs are deploying DSL at a significant rate, but their motives are not necessarily obvious. Are they encouraging DSL because they can charge more for it and at the same time reduce the number of “free” local phone calls that people make to ISPs? If local phone calls prices were de-regulated, would broadband survive in the ILECs? So now, it seems that there is little left of the old “public” network. Long distance is highly competitive, although it is clear that public interest and national security would prevent the failure of certain carriers. Meanwhile local access seems to be moving more towards deregulation and/or monopoly abuse every day. Mobile Networks The other big change in telephony has been the growth of mobile cellular systems. These of course have been wildly successful, and have grown with a regulatory expectation of increasing competitiveness. Of course, in early mobile networks, almost all calls either started or terminated in the older fixed network, so the basic service had to be compatible with the customers’ expectations of a voice call, both technically and economically. This has is turn shaped the mobile industry to be a lot like the fixed telephone industry. For - 4 -
5. voice calls, whenever a call leaves the mobile carrier it can be “dumped” on the old voice network and thereby inherit the broad connectivity of that network. As new services evolved specific to the mobile phone, like SMS (short message service, aka texting), carriers discovered the need to interconnect directly with each other to deliver users’ traffic. The history has been that carriers resist the interconnection with their competitors until users are hugely dissatisfied, then they finally connect, then they discover that volume surges up hugely. However, this interconnectivity is deliberately limited – users can send to users in other networks, but they cannot access competitive services in other networks. The current situation in the U.S. is still not quite there; users can send SMSs to other networks, but the receiving users receive them in a way that they cannot reply to easily! In spite of competition, what has emerged in mobile networks is an oligopoly environment with little differentiation between providers in the types and costs of services, and where the only reliable connectivity is based on the more highly regulated fixed voice network. Through techniques like handset “locks”, lack of number portability, and captive voice mail, carriers make it painful for users to leave their networks. And the dominance of carriers, rather than users, setting requirements for phones means that users can’t get features in their phones that would force more competition between carriers. [WirelessConsumers] One has only to look at roaming charges to see how much a mobile carrier hates their users to use a different network. Not only does the visited network charge a lot, but also the home network inexcusably charges for not using its network! Most recently, the auction of new spectrum for mobile carriers during the craze of the telecom bubble has left many operators with enormous debt obligations for the spectrum. This jeopardizes the previously successful competitive environment because it becomes very hard to raise capital to invest in the network. One expects that the licensing authorities, much to the embarrassment of all concerned, will eventually mostly forgive this debt. The Postal System The granddaddy of all public networks is the postal system, some form of which has been in use for at least 4000 years, in both ancient Egypt and ancient Babylon. [Holtzmann] The advent of written language probably coincides with the earliest “message networks”. The postal network goes through all the same cycles regulation and competition, good and bad interconnection, etc., as any other network utility. The postal system solves not only a messaging problem, but also an addressing problem. How do you sign up for telephone or electrical or water service without giving your postal address? These days, with UPS and Fedex, one might imagine going without postal service, but how would you go without postal addressing? Addressing in public networks tends to be formalized after the fact and take on huge political dimensions as - 5 -
6. people start to understand the full implications. Thus ICANN (the Internet Corporation for Assigned Names and Numbers) seems due to take on at least as much strife as the I.T.U.; whose ugly political battles caused the shapers of the Internet to keep the Internet address governance in a separate organization. The Internet Protocol IP, the Internet Protocol, is the low level technology underlying almost everything labeled “Internet” (The word that is otherwise so devoid of meaning). IP strongly resembles a postal system for computers – there’s a receiver address, a sender address, no ordering required among packets, etc. Like letters, IP assumes that the network does not know or care what is inside the envelope. The roots of IP go back to 1974 when Louis Pouzin suggested the “Catenet” model as a technique for interconnecting packet switching networks.[Pouzin] Vint Cerf suggested this model for the fledgling Arpanet[Cerf] and by 1983 IP had been standardized and deployed throughout the Arpanet, and use was beginning in other networks. The Catenet model is very important in that it unifies the user-to-network abstraction with the network-to-network abstraction and thus allows users to have or be their own networks. In retrospect, perhaps this doesn’t seem too inspired, since pretty much all public utilities behave that way, but certainly it is 180 apart from the dominant telecommunications views, then and now. Besides IP, there were a number of functionally similar competitive protocols in use for similar purposes. Only in 1995, when Microsoft included TCP/IP as a base technology in Windows/95, did the broad industry finally admit that IP was the only “layer 3” protocol that mattered. So why does IP succeed as a network? An ideal network, by most definitions, would have high bandwidth, low latency, no loss or re-ordering, good security, and guaranteed quality of service. But IP has none of these! Sure, it works on top of any technology that may provide those attributes, but the applications that use IP must not assume that any of those attributes are there. Applications that successfully divorce themselves from assuming good things about the network become extremely robust, and the users are happy (enough) with them. Like the postal system, an IP network succeeds because of the definition of its basic service: simple, easy to transport, easy to interconnect, and independent of application. Like the postal system, one should be wary of trusting the network – forgery, theft, hazards, scams, and obscenity are all possible, but are best dealt with by forces outside the network. Again, “common carriage” is the legal model that has evolved over centuries to deal with these types of problems. - 6 -
7. Private IP vs. Public IP Certainly, the early vision of IP was that there would be a single IP “inter-network” which was coordinated within the public IP address space. However, today we find that there are enormous numbers of private IP networks, each controlling their own address space, which may or may not be connected to the public Internet (through NAT – Network Address Translation – products). Metcalfe's Law states that “the usefulness, or utility, of a network equals the square of the number of users.” This seems clear enough when the network allows for any-to-any communications between the users. Unfortunately, NAT forces a division of “users” into clients and servers, and clients lose direct communication unaided by network services. With the dominant applications of the Internet today, consumers tend not to notice what they lose with NAT, and in fact the loss makes their security problems much easier to manage. Services, a.k.a. content providers, clearly cannot live with NAT as the only network access. From an economic point of view, NAT can be viewed as an anti-competition tool. NAT makes it extremely difficult to connect to more than one network provider, and makes it hard for a user network to transition to a peer network. Today, if we were offered a clean slate, NAT would not be necessary, but for one reason - addressing. Security tools and technology have improved – both end-to-end and firewalls. Nobody really wants a disjoint network, because of the importance the Internet has taken on. The obstacle now is the limited 32 bit address space of IPv4, which, curiously, was not the reason NAT was deployed, but will be the reason that it stays. Later, we will hear more about IPv6, which solves this one key problem with IPv4. User mobility is another phenomenon that is driving the collapse (or irrelevance) of private networks. Users have a strong preference for having a single “mode” of operation as they move themselves and their devices among work, home, hotel, and other locations. The set of services typically available directly on a corporate LAN can be difficult to provision for remote use with VPN or RAS tools, and once it works the user must switch modes when he enters and exits the work LAN. But outside of the work LAN, all network environments tend to look quite similar, i.e., public Internet access. So the way to achieve a single modality is to change the LAN to also look like public Internet access! This is even more true as wireless LAN technology spreads and threatens all notions of LAN privacy because of its weak security and uncontrolled availability. Bit Pipes and the Borg Because IP assumes so little from the underlying technology, it is possible to run IP, and therefore any Internet service, over any “bit pipe”. It may be ugly, slow, unapproved or illegal, but the pressures to do so will be substantial. - 7 -
8. There is a continuing history of network providers and equipment manufacturers who are in denial about IP and the Internet. First, they build new networks that they hope to be better, faster, cheaper or more attractive than the Internet in some dimension. Upon deployment, they discover that someone starts running IP on top their nice new network and then the Internet Borg extends to include the new network and turn it into just another piece of the Internet. Where are ATM, X.25, Token Ring, Appletalk, etc. today? Resistance is futile. Of course, network providers hate the notion of a bit pipe, because they lose visibility into the nature of the traffic, services, users, etc., and therefore lose the ability for differential pricing. In a competitive environment, this means you have competition providers of a commodity, so price drops to a level that may not support continuing business. However, the bit pipe is exactly what the users want, because of the value of being simply connected to everyone else with unimpeded services. And it is sheer hubris for the providers to think that they can offer every service the users may want, or every terminal, or every application they may need. The value of the network is precisely in the serendipitous economic value that it creates. Economists speak of network externalities in an attempt to quantify this value – the value is external to the network, so the network provider in unlikely to have any control or even understanding of that value. This is especially true with the Internet because no one network provider controls a significant part of the Internet. So, we have the “Network Paradox”, wherein what the users & the economy at large need is essentially impossible to offer as a normal business [Isenberg], and where every attempt to specialize the network with added services creates unwanted dis-services [Cheshire]. But is this not the case with all public utilities? Users and Uses Metcalfe’s law speaks of the number of users of a network, but determining who or what the users of a network utility are can be quite frustrating. Who are the “users” of the railroad network? Are the engineers on the trains the users? If the train is full of little plastic toys headed for McDonald’s, is McDonald’s the user? Are the kiddies who get the toys the users? The laborers who made the toys? The landfills that eventually receive them? Is it more proper to speak of a toy network, a cardboard box network, a shipping container network? As utilities grow, the uses of the utilities become unfathomably complex, and many interdependencies grow between utilities. A reason that utilities must strive for great reliability is the complete unpredictability of exactly what would suffer if the utility fails. Nevertheless, the service offered by any one utility remains quite simple. Utilities live on for decades and centuries. Utilities don’t “converge” with other utilities. The telephone networks bought up telegram networks, yet there was always a difference from the user point of view, and most of the facilities remained separate. Telegram - 8 -
9. service didn’t completely die out in the U.S. until after e-mail was available; and telegrams continue to be widely used in parts of the world where e-mail isn’t. Coincidence? Or are these really the same utility with a technology update? So what kind of utility is the Internet? Is it naturally part of telecoms – the worldwide telephone industry? If so, where does this leave cable Internet? Or is it just a big user of telecoms, the way that the postal system is a big user of the roads? And who, or what, are the users of the Internet? A casual definition of the Internet might include many elements that are not IP connected, but let’s look just at IP. The endpoints of an IP network are computers, mostly. But things are changing in a hurry. There are many elements in the network – switches, routers, name servers, caches, etc., which also use the network. There are enormous numbers of new “appliance” products that are becoming connected – printers, cell phones, IP phones, MP3 players, digital video recorders, web cams, scanners. Industrial applications abound with all kinds of sensors, power & thermal management controllers, etc., becoming IP connected. And even the definition of computer is changing to include networks – clustered, distributed, and virtual computing solutions are everywhere. Microsoft has announced that they’ll use IPv6 to support Bluetooth, and they’ll use Bluetooth for keyboards and mice. So how many IP addresses will a single person use at any time? It seems clear that almost any processor, embedded or not, can benefit from IP connectivity. With the increasing capabilities of VLSI and the increased understanding of networking, it is now conceivable to build network endpoints that are not processor based at all. Moore’s law [Moore] marches on with no end in sight, so at what point does it become trivially cheap to put networking on every chip? I hope the sophistication of security methods can keep up somehow! Now we see who the ultimate user of the Internet is – the chip! Silicon wants to be connected! And there will be many forms of networks to connect these chips which will have nothing to do with “telecoms”, yet would definitely want IP connectivity. Think of the enormous growth yet to come – who cares if half of U.S. adults are online already? Only a tiny fraction of a percent of the chips are! Already, in the Internet backbone, non- interactive traffic is in the majority. Search engine crawlers, backups, forward caching, automatic updates, etc. is taking over the Internet. Silicon has a lot to say, and it can say it really fast. Quality of Service Telephone service providers have always wondered how the Internet could succeed without Quality of Service (QoS). It’s easy: silicon is very, very patient. But seriously, QoS concerns are highly dependent on both the basic services provided as well as the business models that are expected. - 9 -
10. In the telephone system, the ability to place and receive calls – being connected – is a service of substantial value separate from the value of any particular calls. How do you feel if you pick up a phone and there’s no dial tone? Similarly, in IP networks, the ability to send and receive packets – being connected – is a fundamental service separate from any particular packets. Unfortunately, when dial-up users use IP, they don’t have the ability to receive packets unless they first place a call – in which case they lose the ability to receive calls! But yes, IP is tuned to sending packets that way that the telephone network is tuned at delivering calls. So, although the IP QoS mechanisms may not be terribly appealing to voice users; it’ll certainly work better than using circuit switched networks for bursty data. Many utilities provide differentiated services – like Express Mail, or redundant power lines, but these tend to all be defined relative to the most basic service, and the vast majority of users are content with the basic service. For many years, network operators have been emphasizing their need for “multi-service” networks so they can consolidate their transport and operations and offer additional kinds of services. While there is nothing inherently wrong with this goal, it is completely inwardly focused, instead of focused on customer needs. From a user point of view this is “convergence” which usually results in confusion. A user expects simple universal services and simple universal interfaces. When interfaces change just to achieve convergence with no inherent added value, users are uninterested. Consider making voice calls from a PC, or browsing the web from a telephone. These may be fine with the network, but the interface is so different that the users are put off. A network utility needs universality to succeed, and universality requires simplicity. Of course convergence does happen, but not in a visible way. The SONET network has emerged to carry both voice and data; Voice over IP is used quite a bit, but the users don’t know it, because the interface (telephone) is preserved. From a user point of view, quality of service is simple: does the basic service meet the expectations? From the provider point of view, the question is how to drive the users to pay for more than just the basic service. Crippling the basic service should not be the answer. Business Models - Summary To summarize the big problem facing the Internet, it is that the connectivity presumed by the Internet meets the classic definition of a network utility, but that there is little investment in the utility aspects of the Internet due to the regulatory and competitive climate. Connectivity is the fundamental service of the Internet, yet it is connectivity that suffers first when network providers compete for users and services. - 10 -
11. The solution to this problem is to create a global utility for IP connectivity, and to have the public interest well represented in managing this utility. To correct today’s Internet industry to conform to this new model would be a huge and complex undertaking meeting resistance from all sides – economical, political and technical. Yet for those who think this is impossible, I propose the following: quit using IP! I believe quitting IP would be even more difficult because the providers that don’t wish to offer the universal connectivity of IP also benefit greatly from its serendipitous qualities. So we have two steps to consider – defining this new utility, and then finding a way to have it come into being. Let us consider the definition first. Defining a Utility Utilities can be recognized by the simplicity of the service they offer, and by the ubiquity of the interfaces that access the service. For water, the service is water with pressure and cleanliness standards and the interface is a faucet. For electric power, the service includes a definition of volts, amps, and frequency, and the interface is the wall or lamp socket. Also important to the service definition is the assumption that pricing is simple and doesn’t change rapidly. In fact, a good service and interface definition is a pre-requisite to a successful utility, and can determine how well the utility scales technically and thus what scales of economies are achieved in building it. The global power grid is a hodgepodge of different voltages, frequencies, and sockets, so the equipment markets are limited (somewhat) in achieving efficiency. The interface definition for the IP utility is clear: Ethernet. From 2 Mb wireless in the home to 10 Gb fiber in the backbone, Ethernet is the ubiquitous physical manifestation of IP. It is already a standard of global reach, and philosophically compatible with IP – for instance, each Ethernet port can connect either a single terminal or another network. We, know, however, that the service definition of Ethernet does not scale to millions of nodes, so it is not the correct choice for a service definition. The obvious service definition for the IP utility would be simple, global, routing of public IP. We know already that IPv4 is limited in scale, so IPv6 is a better technical choice. The beauty of choosing this transition to IPv6 is that it defines a new network instead of attempting to restructure the current ones; yet the transition and interoperability scenarios are still simple and well defined.[RFC2893] IPv6 defines the syntax of the service, but the IP utility service must be a common carriage service to allow universal semantics. Common carriage requires the provider to be uninterested in the contents of packets, and it requires excellent interconnections with other providers. Job number 1 is to get packets to other networks; there can be no preference given for sources, destinations, users, or services resident in one’s own network. In return for the duty of common carriage, the users must not expect the provider to be responsible for the content of packets. - 11 -
12. The pricing of the IP utility will be based on fixed connectivity charge (may be large) with a variable usage charge (may be zero). The usage-charging model must allow users to evolve to providers; so a symmetric pricing model is needed, i.e., either sender pays or receiver pays. Since one has little control over what is received, sender pays makes more sense. Roaming users obviously don’t have fixed connectivity, so only usage charges should apply. (The IP utility is inherently “always-on”, so dial-up and its related billing problems don’t apply) Interaction with Current Utilities Due to the high costs and low margins of building a network utility, the IP utility will most likely require either public ownership or the awarding of monopoly franchises to attract capital. The current telephone and cable industry players will most likely cry foul due to their current investments in Internet infrastructure, and their perceptions of their regulatory mandates. The regulatory contract with telephone companies is to deliver a service defined by common carriage of voice or digital, calls or circuits. Routing packets is clearly not part of this service, although it certainly can make good uses of circuits for connectivity. But routing extends far beyond the boundaries of telephone networks and thus cannot be assumed to be a subset thereof. Cable television delivers television signals as a primary service; the technology isn’t really capable of common carriage (of TV) so as a result they’ve never felt impeded in exploiting IP on their networks. However, the closed heritage tends to make the cable operators unsympathetic to the common carriage desires of IP users. The well known “open access” issue of cable networks is attacking the wrong problem. Yes, users need access to any Internet services, but the cable medium cannot support an arbitrary number of transport channels like the phone system can. Thus, Internet transport should be separated from the ISPs and placed in the cable network. But this is not an attractive business for the cable operators. Of course, the same facilities that telephone and cable operators use to provide their current services could also provide the IP utility. However, they must be willing to abide by the service definition of the IP utility. They may also discover that it becomes more efficient to offer their old services by using facilities of the new IP utility! Likewise, electric or any other utility could use or compete to provide the new IP utility. It is interesting to note that some mobile telephone operators are moving to a kind of utility model as well. Because of the clear expense and overlap of having multiple network providers each building networks of the same technology, operators have begun to agree on “network sharing”. But to the extent that the sharing agreements are opaque to users and regulators, it could be at the expense of the competition assumed when the regulators granted multiple licenses. Therefore, additional regulatory pressure may be needed to protect the public interest, or perhaps we should just formalize network sharing as regulated network utilities! - 12 -
13. Building the IP Utility So how can we begin to build the IP utility? Remember, utilities are born of local needs and desires; if their service definition is good enough then they can continue to grow and coalesce until they attract capital and more formal public oversight. The opposite approach of attempting to mandate from the top-down is doomed to be de-railed by entrenched interests and current regulatory structures. And, of course, the IP utility can be rightfully criticized as anti-competitive, socialistic, and too skewed in favor of particular technology. But as more of us tire of competition and desire stability, as users begin to understand the importance of connectivity, and, if IPv6 is truly necessary, sufficient, and desirable, then the IP utility will be unstoppable. The Institute for the Promotion of the Internet Protocol Utility (IPIPU) will be a non- profit association of users of Internet connectivity and services that will own the definition of the Internet Protocol Utility TM and will certify providers as conforming to the service definition using the IP-OKSM seal of approval. Users, then, can begin to place concrete demands upon their providers, they can easily verify whether providers offer an IP-OKSM service, or they can follow the Internet Protocol Utility TM definition and build their own networks, knowing they can be easily interconnected with others. Two clear requirements for the IP Utility are that native IPv6 is offered at all interfaces, and that the IPv6 is connected to the global IPv6 backbone. Thus, the IP Utilities are inherently interconnected from the start, so the “network effect” can take hold. In the early days, most traffic in the utility will be IPv4 over IPv6, so older ISPs and applications still function, and the interconnect will most likely be IPv6 tunneled over the IPv4 Internet. A long term view of progress needs to be maintained, because the network effect will not be noticeable until there are substantial numbers of endpoints. The IP Utility is certainly no panacea for all the problems of the Internet. But if we can focus on this one narrow, yet fundamental, aspect of the Internet, we can really make a difference. The IP Utility encourages continued innovation in technologies beneath IP, and in applications above IP. IPv6 is so simple, and the benefits of interconnection so huge, that I doubt very much that entertaining alternatives would be fruitful. Network Rollouts Let’s take a look at some ongoing network rollouts that could benefit from the concept of the IP Utility, and that are examples of the bottom-up activities that, if interconnected, could become a widespread IP Utility. The City of Palo Alto Utilities department has been conducting a trial of fiber-to-the- home. Yet the service definition is quite fuzzy – are they targeting high speed Internet, or trying to replace the incumbent telephone and cable providers? Yes, the technology could accommodate these, but how can the service thrive and survive without a focus? And it was difficult to find an ISP to connect the city users to the Internet, because there was no - 13 -
14. exclusivity offered – capital investment will shy away when there are no barriers to competition. http://www.cpau.com/fth/ The recent explosion of Wireless ISPs offering IEEE 802.11 wireless Ethernet service in “Hot-Spots” would benefit greatly from the concept of the IP Utility. Instead of each provider putting in their own infrastructure and hoping for non-interference due to the limited range of 802.11, a coordinated effort could provide better coverage, less interference, easier roaming, aggregated backhaul, etc. But there would need to be a trusted 3rd party offering the IP-OKSM service so that the WISPs could still be competitive. In downtown Tallahassee, FL and Pittsburgh, PA there are community access wireless LANs – the benefits are high and the costs are small for cities that can use their own properties for antennae sites. The Click! Network operated by the city of Tacoma power utilities offers an alternative to cable TV, but also a wholesale Internet connectivity service very much like the IP Utility. http://www.click-network.com/ The Ruby Ranch neighborhood of Summit County, CO got together to create their own DSL based Internet access utility. Grant County, WA has over 2000 fiber-to-the-home customers running on their Zipp Network. Community networks may work in one community, but how can they scale to wide areas without depending on commercial ISPs? The Northwest Open Access Network leases dark fiber and is building quite a large backbone connecting various community utility districts in Oregon, Washington, and Montana. Many states & countries have dedicated high speed wide area networks for research & education. These could form the backbone of an IP Utility for universities – a different kind of community. Many of the communities we mention are what would have been considered by commercial ISPs and carriers as quite unattractive markets. Yet, because they got together and connected themselves they end up with affordable and state-of-the-art networks. Meanwhile, much of Silicon Valley goes without broadband. So, to paraphrase radio commentator Scoop Nisker, If You Don’t Like the Internet… Go Out and Make Some of Your Own! References Adelman, K. 2002, http://www.solarwarrior.com Cantelon, P.L. 1993, The History of MCI – The Early Years. Dallas: Heritage Press - 14 -
15. Cerf, V. 1978, The Catenet Model for Internetworking, http://www.isi.edu/in-notes/ien/ien48.txt Cheshire, S. 1996, For every Network Service there's an equal and opposite Network Disservice, http://www.stuartcheshire.org/rants/Networkdynamics.html Farrer, T.H. 1902, The State in its Relation to Trade. London: Macmillan F.C.C. 2002, Telephone Subscribership Report [Data through Nov. 2001] http://www.fcc.gov/.../subs1101.pdf Gilligan, R. and Nordmark, E. 2000, Transition Mechanisms for IPv6 Hosts and Routers, RFC-2893 Holzmann, G.J and Pehrson, B. 1995, The Early History of Data Networks. IEEE Computer Society Press Isenberg, D. 2002, SMART Letter #68, http://www.isen.com/archives/020317.html Isenberg, D. and Weinberger, D. [Draft 2001], The Paradox of the Best Network. http://www.netparadox.com Moore, G.E. 1965, Cramming More Components onto Integrated Circuits, Electronics Magazine Vol. 38 No. 8 (http://www.intel.com/research/silicon/moorespaper.pdf) Newbery, D.M. 2000, Privatization, Restructuring, and Regulation of Network Utilities. The M.I.T. Press Pouzin, L. 1974, A Proposal for Interconnecting Packet Switching Networks, Proceedings of EUROCOMP, Bronel University Wireless Consumers Alliance, Complaint filed on April 5 2002 in the U.S. District Court for the Southern District of New York. http://wirelessconsumers.org/Anti-trust Complaint.pdf - 15 -