Telecommunications Policy 34 (2010) 405–416 Contents lists available at ScienceDirect Telecommunications Policy URL: www.elsevierbusinessandmanagement.com/locate/telpolCritical resource: An institutional economics of the Internetaddressing-routing spaceMilton Mueller a,b,Ãa Syracuse University, School of Information Studies and Technology, USAb University of Delft, the Netherlandsa r t i c l e in fo abstract This paper links the analysis of IP address policy to the established vocabulary andKeywords: concepts of institutional economics. Internet addressing and routing are usuallyIP addresses discussed in technical terms, yet embedded in this highly technical discourse are aInternet governance number of critical economic concepts, such as scarcity, externalities, common poolInstitutional economics resources, tragedy of the commons, and conﬂict over the distribution of costs. To solveRegional Internet Registries these problems, governance institutions native to the Internet have evolved. Yet despiteCommon pool resources the centrality of addressing and routing to Internet governance, there is very littleIPv6 research literature that bridges economic, institutional and technical discussions of IPInternet protocol version 6 addressing and routing. This paper connects the techno-economic discussion to analysisARINRIPE of institutions and governance arrangements.APNIC & 2010 Elsevier Ltd. All rights reserved.1. Introduction ‘‘Address space exhaustion is one of the most serious and immediate problems that the Internet faces today.’’1 Thereader might assume that the statement was drawn from a recent Internet Engineering Task Force (IETF) meeting or acurrent trade journal. Surprisingly, it was published in May 1992—in the very earliest stages of the Internet’s opening topublic use (Wang & Crowcroft, 1992). It reveals the persistence of address space management as a policy issue in Internetgovernance. Today, almost two decades later, there are again worries about the exhaustion of the Internet address space(Hain, 2005). This is fueling a drive to migrate to a completely new but incompatible Internet protocol with a much largeraddress space: Internet Protocol version 6 (IPv6). Given the dominance of the Internet in today’s communicationsenvironment, this (hoped-for) migration and the associated addressing-routing issues are among the most importantissues in contemporary communications policy. Addressing and routing are usually discussed in technical terms (e.g., Meng et al., 2005; Meyer, Zhang, & Fall, 2007).Embedded in this highly technical discourse, however, are a number of critical economic issues, such as managing scarcity,handling externalities and switching costs, avoiding a tragedy of the commons, or negotiating the distribution of costs. Tosolve these problems, governance institutions native to the Internet have evolved. Yet despite the centrality of addressingand routing to Internet governance, there is very little research literature that bridges economic, institutional and technicaldiscussions of IP addressing and routing. The premise of this paper is that the techno-economic discussion must be joined Ã Correspondence to: 307 Hinds Hall, Syracuse University, Syracuse, NY 13244, USA. Tel.: + 1 315 443 5616. E-mail address: Mueller@syr.edu 1 This paper evolved out of consulting work commissioned by the International Telecommunication Union. The author wishes to acknowledge thecooperation and support of Xiao Ya Yang and Richard Hill of ITU-SP in particular. The content of this paper reﬂects the views of the author alone.0308-5961/$ - see front matter & 2010 Elsevier Ltd. All rights reserved.doi:10.1016/j.telpol.2010.05.002
406 M. Mueller / Telecommunications Policy 34 (2010) 405–416to a critical analysis of institutions and their resource governance arrangements before it can provide an adequate basis forpolicy. This paper provides a basic analysis of the institutional economics of IP addressing and routing, with an emphasis on theway allocation and assignment policies are being applied to the new IPv6 address space. The discussion is divided into foursections. Section 2 analyzes the techno-economic characteristics of IP addresses and their interaction with routing inabstract terms. Section 3 analyzes the institutions that have evolved to manage the resources. Section 4 focuses on the IPv6address space, and analyzes the standards, policies and fee structures that are starting to be applied to IPv6. Section 5brieﬂy identiﬁes some of the open issues regarding how IPv6 might raise new policy issues and alter some of the economicand institutional pillars supporting the current address management regime. The primary purpose of this paper is to linkthe analysis of IP address policy to the established vocabulary and concepts of institutional economics. The existing literature in this area is thin. The most comprehensive contribution so far is DeNardis (2009), whichexplains the role of the address space in Internet protocol development and documents some of the history and hopessurrounding the development of IPv6. An OECD report (Perset, 2007) provided a useful overview of the debate over addressscarcity and IPv6 but, like many government agency reports, lacked theoretical grounding and had to be vetted by partieswith vested interests in its analysis. A policy analysis by Mueller and Kuerbis (2008) ﬂags some of the high-levelinstitutional issues raised by IP addressing. Other recent literature has focused on the heated debate over the possibility oftrading increasingly scarce IPv4 addresses (Lehr, Vest, & Lear, 2008; Mueller, 2008; Edelman, 2009; Hofmann, 2009; Dell,2010). None of these papers, however, link the techno-economic characteristics of the resource to a theoretically informedanalysis of the institutions that have developed to manage them.2. Techno-economic features of IP addressing/routing2.1. The address space The Internet protocol creates a virtual resource, the IP address space, of ﬁnite dimensions. It is similar to a block of radiofrequencies dedicated to a speciﬁc service by technical standards. The address space size is ﬁxed by the technical standardsdeﬁning the Internet protocol. IP addresses are scarce in the strict sense that economic theory deﬁnes scarcity: it is notpossible for all of us to have all of the addresses we would like at zero cost. As a virtual resource, they are not ‘‘consumed’’or used up when put into production; rather, they are occupied just as radio spectrum is occupied. When the occupation ofone party ends, the resource could be available for others to occupy. In Internet discourse, the reservation of an addressblock to a single network or administrator for direct use is called an ‘‘assignment;’’ the occupation of a larger contiguousblock of addresses by an intermediary for assignment to others is called an ‘‘allocation’’.2.2. The routing space This relatively simple picture of a ﬁxed resource space breaks down when routing is brought into the picture. Routing isthe process that guides the movement of Internet protocol packets from their origin to their destination. The possession ofIP addresses is signiﬁcant and valuable only insofar as they can be used to route packets on the public Internet. Likewise,routing needs globally unique IP addresses in order to deﬁne routes that can guide packets to their destination. The policyproblem is not simply one of efﬁciently allocating IP address resources; it is also the efﬁciency and scalability of Internetrouting. Analysis of the resource space is complicated by the interdependence of routing and addressing, and the need totrade one off against the other. Thus, the resource with which this report is concerned is referred to as the ‘‘routing-addressing space’’ rather than solely as ‘‘IP address resources’’, while recognizing that addresses and routes aredistinguishable parts of that resource space. Routing is the automated process that directs Internet protocol packets from their origin to their destination. On theInternet, the scaling properties of routing operations and equipment are affected by the way address blocks are distributedamongst networks. IP addresses can be described as part of the language that routers speak to each other. Internet routingprotocols consider the IP address to be composed of two parts: the address of the network (the preﬁx) and the address ofthe connected computer (the host). Routing through the Internet is based on the network portion of the address. A routerstores its best and alternate routes for each preﬁx and uses this information to construct a forwarding table (the routingtable) that controls the movement of each incoming packet to the next hop in its journey. Routers also transmitannouncements to other routers about the address preﬁxes to which it is able to deliver packets, and this information isincorporated into the tables of other routers. Currently, interactions among routers are based on an Internet standardknown as the Border Gateway Protocol (BGP).2 Thus, routers are engaged in constant, automated conversations with eachother that exchange network preﬁxes and other information to keep every router informed about how to reach hundreds ofthousands of other networks on the Internet. BGP exchanges among independent network operators are rife with externalities (see, e.g., Huston, 2001, p. 13; Cowie,2009). There are two major routing externalities: (1) the rate of change in one network’s routing announcements and (2) 2 Rekhter and Li (1995) and Rekhter, Li, & Hares (2006).
M. Mueller / Telecommunications Policy 34 (2010) 405–416 407the size of the routing table. Rapid change in routing announcements can increase the processing load of other networkrouters all over the world. And if no policy limits are placed on the number of route advertisements, it is possible that thesize of the routing table in the largest and most interconnected Internet service providers (known as the default-free zone)3could grow until it exceeds the processing power of their routing equipment. This is an externality because when onenetwork adds announcements of many different fragments of an address block to the routing space it does not make itsown operations much more expensive (and may make them less expensive or gain revenue sources as a result), but whensuch behavior is repeated across many other actors it makes the size of the table in the BGP routers used by ISPs larger andlarger, making routing equipment more expensive. Some have described this problem as a tragedy of the commons(Huston, 2001; Rekhter, Resnick, & Bellovin, 1996).2.3. Route aggregation During the early-mid-1990s, as the Internet took off as a public medium, the number of preﬁxes listed in routing tablesbegan to grow at an alarming pace. Some felt that the scaling problem threatened the viability of the Internet. In anattempt to control this problem, the IETF and ISPs introduced provider-based address aggregation, a hierarchical approach toaddress allocation that strives to minimize direct assignment of IP addresses to end users. It gives network serviceproviders (ISPs) larger address blocks, and encourages most other Internet users to use sub-allocations from these largeraddress blocks, which are aggregated by the intermediary provider into a single route announcement. This minimizes thenumber of entries in the routing tables, and also reduces the amount of trafﬁc exchanged among routers—two veryimportant economic efﬁciency beneﬁts. But provider-based route aggregation has two other economic consequences. First, it increases end-user switching costsin the market for Internet services. The customers of ISPs cannot take their address blocks with them when they changeservice providers. Second, it militates against trading, subdividing or other uncontrolled transfers of address blocks amongend users and organizations. The ability to move unused address blocks or portions of a block from one user to anotherwould greatly increase the efﬁciency of address space utilization. But such transfers would also require breaking upcontiguous blocks of addresses into separately routed parts, undermining the efﬁciency of routing.3. The governance regime With a basic understanding of the techno-economic features of addressing-routing in place, this section integratesinstitutions into the analysis. The current regime for governing IP address-routing resources is organized around RegionalInternet Registries (RIRs). The ﬁrst address registry was a centralized function performed by U.S. research and militarycontractors. As the Internet grew and became internationalized and privatized, the registry function was delegated tononproﬁts serving different world regions. The rationale for RIRs was ﬁrst set out in RFC 1174 (Cerf, 1990). There are nowﬁve RIRs.4 In the early 1990s, as noted earlier, the Internet was beset by serious scaling problems, involving both routing tablegrowth and address space depletion. In response, the IETF created a mixture of technical and policy adjustments thatestablished aggregation and conservation as basic principles for address-routing governance. Classless Inter-DomainRouting (CIDR) was deﬁned and implemented as a way to make more efﬁcient use of the diminishing IPv4 address pool(Fuller, Li, Yu, & Varadhan, 1993). The provider-based leasing model approach to routing, mentioned above, was developedaround the same time and codiﬁed in RFC 2008 (Rekhter & Li, 1996). The RIRs’ policies and practices incorporated CIDR andprovider-based aggregation, and their role was formalized in RFC 2050 (Hubbard, Kosters, Conrad, Karrenberg, & Po, 1996).In 1997 the U.S.-based central Internet registry was fully privatized and became ARIN; in 1998 there was partialprivatization of IANA alongside the creation of ICANN. The RIRs created a more formal contractual regime, and feestructures for membership/address blocks were put into place. Thus, by 1997 the regime had converged around the threeprinciples of registration (maintaining the uniqueness of addresses, aggregation and conservation). Aside from growingand becoming more professional, little changed during the next decade. The RIRs are now a mature transnationalgovernance regime composed of nonproﬁt, private sector membership organizations that govern primarily through privatecontract. The membership is composed of autonomous systems; that is, organizations that operate networks and utilizeInternet addresses. 3 ‘‘Default-free zone’’ refers to the collection of all Internet autonomous systems (ASs) that do not require a default route to send a packet to anydestination. A default route is used by a router when no other known route exists for a packet’s destination address. Typically, defaults are used bysmaller networks which rely on larger Internet service providers to ﬁnd the destination. ASs in the default-free zone are generally the largest and mostinterconnected networks. 4 ARIN (established in 1997), but inheriting functions performed by Jon Postel and other government contractors since the early 1980s, provides theregistry/governance function for North America. RIPE-NCC (established in 1991) serves Europe and the Middle East. APNIC (established in 1995) is theinstitution for the Asia-Paciﬁc region (which includes India). LACNIC (established in 2002) serves the South American continent and the newest RIR,AFRINIC (established in 2005), serves the African continent.
408 M. Mueller / Telecommunications Policy 34 (2010) 405–416Table 1Classiﬁcation of goods..Adapted from Ostrom, 2005, p. 24 Difﬁculty of excluding users Subtractability of use (rival occupation or consumption) Low High Low Toll goods Private goods High Public goods Common pool goods3.1. Common pool governance? Various discussions of IP addresses, including statements by the Internet address registries themselves, assert thataddresses are ‘‘public resources’’. ‘‘Public resource’’ is an unscientiﬁc term, however; its meaning varies and the differencesin usage often reﬂect political agendas.5 Institutional economics provides a more precise and useful distinction betweenfour broad classes of goods: public goods, private goods, club goods, and common pool resources. The distinctions hinge onthe degree to which resources are rival in consumption (i.e., one person’s consumption does not prevent anyone else fromusing it) and excludable (i.e., the degree to which an owner or appropriator of the resource can prevent others fromappropriating it.) With public goods, consumption is nonrival and exclusion is difﬁcult or impossible. With private goods, consumption isrival and exclusion is relatively easy. Thus for private goods, market allocation is the most common option (although it isusually shaped by government regulations or subsidies in some way). The resource is allocated by means of exchanges ofproperty rights among private owners; prices go up as supply goes down and conservation incentives adapt accordingly.Resources do not fall unambiguously into these categories, and their status can change. Encryption technology, forexample, made it possible to exclude owners of a radio receiver from access to a broadcast signal, transforming it from apublic good situation to a toll good. Common pool resource management regimes are responses to a unique set of economic conditions. Consumption isrival, and thus private appropriation must be rationed or limited in some way. But if exclusion is too costly, orinterdependencies among users of the resource make it too difﬁcult to deﬁne and enforce tradable property rights, then aprivate market may not work. An example of a resource which faces difﬁculty of exclusion is a school of ﬁsh in the ocean,which cannot easily be fenced in. In these cases collective governance rules can take the place of market prices as theallocator of the resource.3.2. Rival consumption and exclusion If the framework in Table 1 is applied to IP addresses, it is not immediately apparent why IP addresses are currentlygoverned on a common pool basis. Clearly, address assignments and allocations are rival. If one network occupies anaddress block, another network cannot also use it on the global Internet without creating conﬂicts in routing. Addressesmust be exclusive and globally unique to function properly. Exclusion is also possible, although not in as straightforward a way. When a user appropriates a natural resource like amineral or a ﬁsh from the ocean, the act of possession and consumption by one person physically prevents others from alsoappropriating the resource. The act of assigning numbers to host computers on a network, however, does not by itselfprevent anyone else from also assigning the same numbers to their own hosts. To maintain the exclusivity of assignmentsrequires collective action in the form of a registry accepted by Internet network operators as an authoritative coordinationinstrument. IP address registries meet this need. They keep track of which organizations are using which address blocks and ensurethat the allocations are exclusive. But the registry’s ability to maintain exclusivity depends heavily on its acceptance andrecognition by Internet service providers as a guide to their routing decisions. To enforce exclusive assignments of IPaddresses, ISPs must refuse to route trafﬁc to another network’s address if it is not registered as the legitimate holder ofthat address. The governance power of the registry, in other words, depends heavily on the assent and participation of theISPs who operationalize routing. An address registry cannot simply take back an assignment or allocation in the way thatphysical property is repossessed; it can only eliminate a party from the list of authorized holders and hope that ISPs adjusttheir routing practices accordingly. Thus while exclusivity is more complicated, it is still clear that maintaining exclusivity in the occupation of IP addressesis perfectly feasible. Laws, regulations or contracts could give registered address block holders a right to take legal actionagainst ‘‘trespassers’’ who occupied address blocks that conﬂicted with those of a registered user. This leaves a keytheoretical puzzle: why don’t IP addresses fall into the private goods quadrant of Ostrom’s matrix? 5 It can be used to mean state-owned resources; or privately owned but state-regulated; or it can simply mean a resource that is publicly shared; or itcan mean a public good in the strict economic deﬁnition described below, or it can mean an essential facility as the term is used in antitrust law.
M. Mueller / Telecommunications Policy 34 (2010) 405–416 409 The answer is that the choice of a common pool regime and its restrictions on address ownership and trading weremotivated neither by difﬁculties of exclusion nor by the absence of rival consumption. Rather, common pool governanceemerged because of the decision to use address allocation as a tool for enforcing route aggregation. Concerns aboutaggregation prevented market trading of addresses, and in the absence of a market price system, some other method had tobe used to limit appropriation of scarce number resources. To put it differently, participants in the regime are as concernedabout conserving routing table slots as they are about conserving address blocks, and the interactions of routingannouncements and the structure of address block allocations pushes the industry into a common pool arrangement.3.3. Appropriation limits In the absence of tradable property rights, the existing IP address regime rations and conserves the resource byleveraging the IP address registry function. The registry is more than a mere coordinative device; it also acts as agatekeeper to the address space. The grant of exclusivity and legitimate title is linked to administrative limits on the size ofthe address block any appropriator can claim. Appropriation limits are based on administratively-established technicalcriteria, known in the industry as ‘‘justiﬁed needs assessments.’’ Applicants for address space submit network plans andinformation about utilization levels; the RIRs review these plans and award address blocks accordingly. In short, this is insome ways a centrally planned economy. This approach to conservation is fueled not only by practical concerns about aggregation; nearly two decades ofcentralized governance by a tightly-knit community of engineers has reinforced a quasi-religious commitment tocommunitarian methods. This has produced a resistance to the idea of owning, trading, or speculating in IP addresses thatgoes far beyond the original technical rationale based on the need for aggregation. Though contested and waning, thisideology runs deep within the RIRs.63.4. Reclamation and reuse The problem of reclamation is an important and neglected topic in address governance. Those who have been assignedor allocated addresses but no longer use them are, in principle, supposed to return them to the common pool to make themavailable for use by others, and this is supposed to be one of the primary efﬁciency beneﬁts of a common pool regime. Butresource reclamation has been a persistent weak point of the current regime. There are few positive incentives to returnunused blocks. RIRs have no contractual authority to reclaim IPv4 resources from the so-called ‘‘legacy’’ holders whoreceived their allocations before the contractual regime was put into place. Even for organizations that contract with theRIRs, their ability to monitor the actual use of addresses and reclaim resources from current holders is weak. The scalabilityof detailed monitoring or auditing of thousands of individual organizations is questionable. Thus, huge portions of the IPv4address space are thought to be unused or underutilized, and, contrary to the communitarian ethic underlying the commonpool regime, organizations are not returning them to the common pool. However, a system of digital certiﬁcates that tiesaddress allocations to speciﬁc organizational identities and can be automatically veriﬁed could change all that.4. The IPv6 addressing-routing space Internet protocol version 6 (IPv6) is the next-generation protocol developed in the mid-1990s by the InternetEngineering Task Force to overcome the projected address shortages of the classical Internet protocol (IPv4). There is stillsome uncertainty about whether the migration to the new IPv6 standard will succeed at all (see DeNardis, 2009, chap. 4;Elmore, Camp, & Stephens, 2008). A complete discussion of that issue is outside the scope of this paper. What is known isthat the only real advantage of the new, incompatible Internet protocol over the established one is its larger address space.With 2128 addresses in the IPv6 space compared to the 232 addresses of IPv4, the new protocol constitutes an enormousexpansion. Practically, the IPv6 address space is almost inﬁnitely large.4.1. Conservation and allocation policy for v6 The groundwork for IPv6 address allocation policy was established by the IETF, which developed and ratiﬁed the newstandard. To facilitate the implementation of the IPv6 standard, the IETF made two important recommendations. First, it released only 15 percent of the available global unicast IPv6 address pool for allocation by the Regional AddressRegistries, and left the remaining 85 percent ‘‘reserved.’’ The reservation was explicitly acknowledged as a hedge againstthe possible need for different, possibly more conservative allocation policies in the future. Second, the IETF escalated the scale of allocation policy. The critical principle underlying IPv6 allocation policy is thatallocations are made to applicants based not on the number of individual IP addresses in a block, but on the number ofsubnets. Subnets are large blocks of addresses which can be used to form networks. The smallest unit of subnet allocation inIPv6 is the /64, a subnet size that contains the capacity for 18.4 thousand trillion (18,446,744,073,709,500,000) individual 6 See the literature on address transfer markets alluded to earlier.
410 M. Mueller / Telecommunications Policy 34 (2010) 405–416IPv6 addresses. Note: this is the smallest unit of allocation. It is contemplated that /64 blocks will be assigned to home usersor mobile phones. The basic unit for making allocations to networks of corporations, schools or other end-user organizations is the /48. A /48 subnet contains 65,536 subnets of the /64 size. The basic minimal unit for Internet service providers is supposed to bethe /32, which contains 65,536 /48 subnets and 4.3 billion /64 subnets. In other words, a /32 contains as many /64 subnetsas there are addresses in the entire IPv4 address space. These decisions are set out most clearly in two key RFC documents:RFC 3177 (Internet Architecture Board, 2001) and RFC 5375 (Van de Velde, Popoviciu, Chown, Bonness, & Hahn, 2008). RFC5375 openly acknowledges the detachment of allocation policy from actual counts of computers or devices on thenetwork.7 Measures of ‘‘host density’’ for IPv6 refer not to the actual number of computers or other devices connected, butto the number of different network ‘‘sites’’ to which a subnet has been assigned (Van de Velde et al., 2008; see alsoHuitema, 1994). While its approach may seem extraordinarily liberal, RFC 3177 enumerated several reasons, many of them attractive,for taking this approach to allocation. Most notably, a standardized, provider-independent boundary between the networkportions of address allocations is supposed to make it easier for users to change ISPs without internal restructuring orconsolidation of subnets.8 Even without a change of ISPs, these standardized boundaries may make renumbering easier.Also, large initial allocations minimize the cost burden and administrative overhead associated with ‘‘needs assessment’’by the RIR; it allows ISPs and their subscribers to grow substantially without any need to return to their ISP or RIR withnew (possibly costly) requests for additional (possibly non-contiguous) address blocks. This is, in fact, a very importantpolicy departure from the RIRs’ IPv4-era policy of increasingly restrictive and bureaucratic needs assessments. Initialallocations are no longer based on ‘‘demonstrated need’’ but on some kind of basic classiﬁcation of the applicant.9 In theirimplementation of these guidelines, most RIRs have proposed /32 as the basic initial unit of allocation to Internet serviceproviders. Before requesting more, the RIRs initially proposed to require that these blocks attain a host-density ratio of .80.Later, the recommended HD ratio was increased to .94.4.2. Does conservation even matter? If the IPv6 address space is so indescribably large, is there a need to worry about conservation at all? Themathematically large size of the address space has encouraged some uncritical complacency. Statements that there aremore IPv6 addresses than grains of sand on the world’s beaches or atoms in the universe imply that the number isinexhaustible.10 However, a closer examination reveals that concerns about conservation still exist and cannot wisely beignored. As the discussion above explained, the basic units of allocation are also extremely large, and the distribution ofsuch blocks will, without question, result in the ‘‘waste’’ of vast quantities of bit combinations that could in theory be usedas addresses. In the intermediate term, at least, very few home networks or small ofﬁce assignees of /48 address blocks arelikely to make use of any but a tiny fraction of the 1,208,925,819,614,620,000,000,000 bit combinations their assignmentcontains. Some may grow into that space or discover new applications of networking that make use of it, but many will not.In justifying its liberal policy the IAB in RFC 3177 contrasted the 178 billion /48 preﬁxes available under their plan with ‘‘10billion, which is the projected population on earth in year 2050.’’ The IETF thus implied that everyone on the planet couldbe given a /48 and there would still be plenty to spare. In an attempt to raise a ﬂag of caution, IBM’s Thomas Narten (2005)noted that giving every human on the planet a /48 would occupy ‘‘fully 1% of the available address spaceyin 50 years,’’which he claimed showed that the address space was ‘‘nowhere near practically inﬁnite.’’ While the IETF statement is optimistic and expansive and the Narten statement is cautious, both calculations revealhow little attention is paid in Internet circles to the economic importance of reclaiming abandoned, unused orunderutilized addresses. Both statements assume that giving a /48 address block to every person on earth is a static, one-time decision that would result in an outﬂow of only 10 billion /48 address blocks after 50 years. But this calculation isoverly simplistic and, as a result, far too low. It is necessary to calculate the ongoing ﬂow of address blocks to a populationthat is constantly adding new members (hence making new allocations) and whose ranks are depleted by deaths (thusabandoning allocations which must either be reclaimed or wasted). Assume that none of the allocations freed up by deathsare reclaimed. In that case, depending on the birth and death rates, the number of address block allocations required togive every member of the human race a /48 increases by a factor of three; i.e., it could consume 30 billion address blocks ormore, not just 10 billion.11 Note that organizational populations have births and deaths, too; that is, ISPs, other serviceproviders, and end-user networks are constantly coming into and going out of existence. Unless there is effective 7 ‘‘The practically unlimited size of an IPv6 subnet (264 bits) reduces the requirement to size subnets to device counts for the purposes of (IPv4)address conservation.’’ (Van de Velde et al., 2008, p. 2). 8 For example, if ISPs consistently hand out /48s to end-user organization sites, regardless of size, then an organization that switches ISPs canrenumber easily by simply changing the network preﬁx in all their site addresses. (Whether ISPs will go along with this remains to be seen.) 9 This approach also allows sites to maintain a single reverse-DNS zone covering all preﬁxes. 10 Europe’s Information Society thematic portal, IPv6: Enabling the Information Society. http://ec.europa.eu/information_society/policy/ipv6/index_en.htm. 11 One must sum the initial allocation to the population and all births over the period.
M. Mueller / Telecommunications Policy 34 (2010) 405–416 411Table 2Network number statistics as of May 1992.Source: Wang and Crowcroft (1992). Available Allocated Percent allocated Class A 126 49 39 Class B 16,383 7354 45 Class C 2,097,151 44,014 2reclamation – which cannot simply be assumed but can only be a product of institutionalized policies and incentives – theeffective utilization rate projected must be increased accordingly. Uncertainty about future use, and historical precedent, creates another reason for caution about the need forconservation. The IPv4 space was also considered to be enormous and practically inexhaustible in the early days of theInternet’s development. For the ﬁrst decade of the Internet’s existence (1982–1992), class-based allocations were made togovernment agencies and private sector organizations participating in U.S. government-funded research and develop-ment.12 Near the end of this period as the Internet gradually opened to larger sectors of society and the world, asurprisingly large amount of the IPv4 address space was allocated. Table 2 displays the numbers. It is apparent from these statistics that the initial allocation policy embedded in RFC 791 and its implementation notonly underestimated overall demand, but completely failed to anticipate the actual structure of demand. Demand foraddress blocks was concentrated in the Class B range, whereas demand for small, 256-host networks was much lower thanexpected. Moreover, classful allocations were structurally wasteful, in that applicants whose need fell somewhere betweenthe sizes of the deﬁned classes had to be given the next highest block size regardless of whether they actually needed orwanted the entire additional increment. The impact of these legacy allocations still lives with us today. Latecomers to the Internet party have been greeted withmore restrictive address policies. A document submitted to APNIC by Millet and Huston (2005) called attention to thepossibility of a more rapid than expected occupation of available address resources due to the capacious initialsubnet allocation policies. ‘‘From a public policy perspective,’’ they wrote, ‘‘we stand the risk of, yet again, visibly creatingan early adopter reward and a corresponding late adopter set of barriers and penalties.’’ This campaign from leadingInternet ﬁgures in 2005 led to a tightening of RIR initial allocation policies. ARIN and APNIC eventually adopted a moreconservative policy on initial IPv6 allocations, offering a /56 rather than a /48 to certain classes of users, and increasing thehost-density ratio required for additional allocations from .80 to .94. From an economic standpoint, there are structural similarities between the classful allocations of 1982–1992 and theIPv6 recommendations of the IETF. The IETF proposes to give out /48s on the basis of one’s status as an organization or a /32 on the basis of one’s status as an ISP, with little regard to the variance in the actual requirements of each end user or ISPwithin that category. But the costs associated with withholding addresses from enterprises unnecessarily must not underestimate. A centuryis a very long time for a communication protocol to last. It is difﬁcult to think of a single electronic communication systemstandard that has survived that long without becoming obsolete. Withholding addresses that could be proﬁtably used nowon behalf of future occupants or applications that may never materialize is also a form of waste. Clearly, conservation or appropriation limits of some kind are still needed—but it is also clear that (if we ever get to an IPv6world) allocations can indeed be much more liberal. The main issue here is one of uncertainty about where is the optimalpoint on a tradeoff curve. This is exactly where economic incentives and institutions might be most useful as a mechanism foradaptation. A critical factor in conservation will be efﬁcient reclamation policies and incentives. Oddly, neither Narten (2005)nor Millet and Huston (2005) nor the IETF discuss the incentive for occupiers to release unused or underutilized addressresources. Providing incentives for more efﬁcient reclamation could extend the life of the usable address space considerably.4.3. RIR fees and IPv6 Address fees are a potentially important conservation tool. Ofﬁcially, the RIRs present themselves as member-basedorganizations and their fees as membership dues that recover the cost of their services. Nearly all of them areuncomfortable with any assertion that they are charging fees for IP addresses. The RIRs do provide important services, suchas maintaining the registry and the Whois lookup, performing ‘‘needs’’ assessments, supporting policy discussionprocesses, and so on. On the other hand, some RIR fees reﬂect a positive correlation between the fee size and the size of amember’s IP address allocations. And it also seems logical, in both equity and efﬁciency terms, that people who occupymore address space should pay more, especially when they are commercially exploiting the addresses. But there is an 12 The earliest address allocations were based on the three classes of addresses deﬁned in RFC 791, the basic document deﬁning the Internet protocol.Class A addresses, which correspond to what would now be called a /8, have a network part of the address that is 8 bits long, leaving room for uniqueaddresses for 16.7 million hosts. Class B addresses (now referred to as a /16) had a network preﬁx of 16 bits, leaving room for 65,556 unique addresses.Class C blocks (a /24) allowed for only 256 unique addresses for hosts.
412 M. Mueller / Telecommunications Policy 34 (2010) 405–416 Fig. 1. Comparative fee structures for IPv6 addresses, the 5 RIRs.important difference between fees as methods of recovering the cost of RIR services and fees as rationing devices toencourage conservation of the address space. The RIRs are currently sitting in an uncomfortable space between the two. Insofar as currency differences make direct comparison possible, there is minimal variance in the fee levels across theRIRs for IPv6 address blocks of size /32 and smaller. The normal fees for /32s among ARIN, LACNIC and AFRINIC areidentical (and are all denominated in USD). The fees for address blocks larger than /32, however, vary widely across RIRs.RIPE-NCC seems to have lower rates (denominated in Euros) and a ﬂatter curve, while APNIC seems to have higher rates(denominated in Australian dollars). But in all cases, the fees charged for IPv6 addresses by the three largest RIRs (ARIN,APNIC and RIPE) get larger as more addresses are occupied above the /32 level. This implies that fees are intended toperform a conservation function, and these conservation incentives seem to be directed at larger ISPs. AFNIC and LACNIC,on the other hand, have very simple, two-part fee structures that only distinguish between ‘‘small’’ and ‘‘large’’ allocations,with the dividing line being the /32. All RIRs seem to discriminate between recipients of /32s and /48s based on technicalassessments of need and do not seem to rely much on pricing to encourage conservation of address blocks at that size.However, a recent fee restructuring at APNIC ties fees more closely to block size. APNIC fees respond logarithmically toincreases in the number of addresses allocated. This policy could constitute an interesting new trend in RIR resourcemanagement. Contributing further to the view that fees serve a policy function, ARIN has discounted their standard IPv6 address feesfrom 2008 to 2011 in an attempt to encourage migration to IPv6. Post-discount, the initial fee for a /32 allocation fromARIN in 2009 is a paltry $562.50 a year. (Remember: a /32 block contains as many /64s as the entire IPv4 address space.) Thediscounted ARIN fees are very similar in size and structure to the RIPE-NCC fees. Similarly, LACNIC has exempted membershaving only IPv6 addresses from paying membership fees until July 1, 2012. It appears that, no RIR has raised fees for IPv4addresses as the free pool approaches depletion (see Fig. 1). Once fees are related to the amount of address space consumed, it is evident that RIR fee structures provide massivevolume discounts to ISPs (with the exception of APNIC’s recently-reformed fee structure). Larger occupiers of the spaceenjoy a discount of about 105 in terms of the price per address. Also, while there is much noise against the idea of ‘‘payingfor address space’’ in some quarters, the fact of the matter is that ISPs use address space as an input into their services, andthen turn around and sell Internet service to their customers, often charging for the ‘‘privilege’’ of a ﬁxed IP address.4.4. IPv6 and routing Initial policy toward IPv6 allocations was heavily inﬂuenced by the scaling problems associated with routing that wereexperienced in the IPv4 space. This is one reason why the initial allocations were proposed to be so large. Large initialallocations would allow ISPs and end-user organizations to announce one network preﬁx for a long period of time, ratherthan expanding through acquisition of additional blocks that might (if they are non-contiguous) require additional routeannouncements (see Fig. 2).
M. Mueller / Telecommunications Policy 34 (2010) 405–416 413 Fig. 2. Discounted ARIN IPv6 fees (10-year period). On the other hand, the larger address size of the IPv6 space means that the routing tables will have to carry biggeraddress preﬁxes. This will increase the demand on the memory of routers. And inherently associated with the vast numberof subnets is a potentially much larger universe of routes. This has given some experts concern. An IETF report noted,‘‘Given that the IPv6 routing architecture is the same as the IPv4 architecture (with substantially larger address space), if/when IPv6 becomes widely deployed, it is natural to predict that routing table growth for IPv6 will only exacerbate the[routing table scaling problems].’’ (Meyer et al., 2007) Another sobering fact is that during the transition period from IPv4to IPv6, many routers will have to support routing tables for both protocols. Until adoption increases, it is difﬁcult to knowwhether, or for how long, the larger initial allocations of IPv6 will outweigh the other factors. But in pure potential, the vastsize of the address space and the vast number of subnets and hosts that could be connected through it means that thescaling problems of 1990–1993 could pale in comparison. There are conﬂicting opinions about the routing problem within the technical community. Huston (2009) claims thatgrowth in the routing table is predictable and falls within the limits of Moore’s Law (i.e., increases in the processingpower of routers will compensate for growth in routing table size), therefore BGP can be extended for the intermediatefuture, at least. On the other extreme, there are those who believe that BGP is inherently fragile and that an adequateresponse to its scaling problems and security issues will require a new routing architecture that separates locators fromidentiﬁers. (O’Dell, 1997; Meyer et al., 2007) In April of 2010, an Internet draft deﬁning Locator/ID Separation Protocol(LISP) was published and started to be used experimentally (Farinacci, Fuller, Meyer, & Lewis, 2010). It is also conceivablethat network address translation and IPv4-IPv6 translation will become stuck in place as a de facto new routingarchitecture.5. Future economic and institutional issues As noted earlier, there is still some uncertainty about whether the migration to the new Internet standard will succeedcompletely. The basic problem is that migrating to IPv6 imposes signiﬁcant costs on organizations and yields nocorresponding immediate beneﬁts. The migration will incur costs: training, weeding out compatibility problems, softwareand hardware upgrades. Moreover, an IPv6-capable Internet service provider will have to keep running both protocolstacks (both IPv4 and IPv6) for some time – many estimate that it will take decades to fully replace IPv4. After all theseexpenditures the end result will be, at best, an Internet that functions exactly as it did before. Aside from the expandedaddress space, IPv6 has few superior capabilities (better support for mobility is one of them).
414 M. Mueller / Telecommunications Policy 34 (2010) 405–416 For large ISPs with growing customer bases, the expanded address space does have some beneﬁt, but for manyorganizations whose current assignments are sufﬁcient, there is no incentive to move whatsoever. Not only are there fewﬁrst-mover advantages to adopting IPv6, for many organizations the rational strategy is to wait for everyone else to go ﬁrst,and only change when one is forced to by the prospect of incompatibility with others. Assuming that IPv6 does take hold, the movement from an environment of address-routing scarcity to oneof address abundance with continued constraints on routing promises to have transformational effects on the policies andinstitutions for critical Internet resource management. Institutional economics suggests two distinct perspectives that can betaken. One emphasizes the role of inertia and path dependency as an explanatory factor in institutional evolution(North, 1990). The other emphasizes the way major discontinuities in supply and demand and in the incidence of costs andbeneﬁts can produce radical qualitative shifts in institutional regimes (Mueller, 2002). In this case both pressures will play anactive role. The large size of the IPv6 address space challenges the very basis of ‘‘needs assessment’’ as the basis for addressblock allocations. As noted earlier, the IETF allocation standards already constitute a major departure from classicalneeds-based methods. Users are assigned blocks more on the basis of their status and the number of ‘‘sites’’ theyhave than on ‘‘demonstrated need.’’ Many holders of allocations will be given address blocks that vastly exceed theirimmediate needs. Only when organizations ask for additional addresses will actual utilization levels be assessed.Currently, a lot of the RIRs regulatory leverage comes from their status as being the only source for new addressallocations. In the IPv6 world, the initial allocations will be so large that RIRs may have less continuing leverage on userorganizations, because they will not need to come back asking for more addresses very often. IP address abundance maymake it more difﬁcult for Internet registries to prevent those holding capacious allocations to reassign all or partof their blocks for use by third parties. Portions of that address space might be leased out to those who want them butconsider the RIR process too costly or bureaucratic. As long as an Internet service provider can be found who will routethat address space, it would be difﬁcult for address assignment authorities to keep track of, much less prevent, suchsub-delegations or leasings. On the other hand, the RIRs could actually strengthen and rigidify their control over address allocation. The RIRs arepromoting a security technology known as Resource Public Key Infrastructure (RPKI). RPKI is a digital certiﬁcate schemethat allows ISPs to authenticate whether someone using an address block was actually assigned that block by one of theRIRs. It also allows the routing paths announced by ISPs to be authenticated as well. The RIRs are taking aggressive steps toimplement RPKI rapidly, and the Internet Architecture Board has announced that RPKI should be organized into a single,hierarchical, global trust anchor, probably controlled by IANA. While this would add some security to what is now a looselyorganized routing system and prevent address hijacking, it would also provide the RIRs with direct control over ISPoperations and their use of address resources. The current regime does not provide this. A fully automated RPKI systemapplied to both address assignments and routing could give the RIRs the ability to disable Internet service providers whodo not conform to their policies. This, along with the issue of who stands at the apex of the trust anchor hierarchy, poses ahost of new governance problems (Mueller & Kuerbis, 2008). The shift to IPv6 also raises questions about the proper geographic scope of an address governance regime (i.e., should itbe global, national or regional?). The regional nature of RIRs is something of a historical artifact, driven more by politicalconcerns than by technical ones. There are tensions between the regional status quo, the technical imperatives ofgoverning the Internet (which lead toward global governance), and political imperatives (which sometimes lead to areassertion of national arrangements). Address abundance undermines the rationale for localized needs assessments, and while the problem of routeaggregation remains, routing externalities respond to global interactions among ISPs and are based on the topology of thenetwork, not on geography per se. Many of the policies and problems associated with IPv4 scarcity also required globallycoordinated responses, such as shifting reclaimed address space from one region to another. So far, the RIRs have not beenable to establish coordinated policies to attack these problems. It is possible that various forms of arbitrage will emergeand take advantage of the differences among the regional IRs’ policies. Furthermore, the possibility of a new routing technology, emerging either from the IETF or from de facto adjustments tothe v4-v6 migration, could lead to major institutional changes. In the past 15 years, route aggregation has provided one ofthe key rationales for the RIRs central management of address resources. Although few RIRs would openly acknowledge it,the rationale for ‘‘needs assessment’’ has been undermined by the abundance of IPv6 resources and by the obviouspossibility of address block trading. RIR defenders have basically lost most of the arguments against market transfers as aform of rationing address resources. The only credible reason they can provide for preventing market transfers and forcentralized coordination of allocations is the need to maintain hierarchical aggregation. What happens to the RIR regime ifa new routing architecture evolves and BGP and its peculiar problems go away? In contrast to these technical pressures, it is evident that some nation-states, working through the InternationalTelecommunication Union (ITU), want to reassert a traditional nation-based governance model for Internet addresses. Arecent report commissioned by the ITU explicitly advocates a parallel system of Country Internet Registries (CIRs) managedby the ITU as a competitive alternative to the existing RIR regime (Ramadass, 2009). Many ITU members prefer a CIRarrangement not because of any real or imagined beneﬁts of competition, but because they are concerned about resourcedepletion and political control. Developing countries fear a replay of the address land rush that characterized the early daysof IPv4; reserving a substantial block of IPv6 address to each national government not only secures their access to the
M. Mueller / Telecommunications Policy 34 (2010) 405–416 415resource but also makes it easier for them to exert policy control over the Internet industry. Critics of this proposal fear thatnational governments would abuse that power to censor, over-regulate or even partition the Internet. They also believethat aggregation requires a globally coordinated approach and do not trust the national governments and ITU to follow theappropriate policies. A number of critical values are at stake in IP address allocation. The future of global compatibility is implicated by theapparent need to migrate to a new Internet standard that is not backwards-compatible. It is not known exactly how liberalor conservative the initial allocation of the new address resources should be, but the decisions made now will have a majorimpact on the costs and accessibility of Internet resources decades from now. The future of the end to end principle isimplicated by the numerous gateways, network address translators and kludges that might be required to extend the life ofthe IPv4 space or to maintain connectivity between the IPv4 and IPv6 Internets. There are also weighty geopolitical andpolitical economy issues involved in the transition. The IPv6 address-routing space can be compared to the opening of avast new continent. It is inevitable that political and institutional authorities would compete over the control of thatresource space. The scalability of the Internet – its continued growth – cannot be taken for granted unless these problemsare solved.ReferencesARIN. (2009, September) Number resource policy manual (Version 2009.4). Herndon, VA: American Registry for Internet Numbers. Retrieved from /https://www.arin.net/policy/nrpm.htmlS.Cerf, V. (1990, August). IAB recommended policy on distributing Internet identiﬁer assignment. (RFC 1174), Internet Engineering Task Force. Retrieved from /http://www.rfc-editor.org/rfc/rfc1174.txtS.Cowie, J. (2009). The adventurous parts of the Internet. Retrieved from Renesys blog /http://www.renesys.com/blog/2009/01/the-adventurous-parts-of-th e-i.shtmlS.Dell, P. (2010). Two economic perspectives on the IPv6 transition. Info, 12, 4.DeNardis, L. (2009). Protocol politics: The globalization of Internet governance. Cambridge, MA: MIT Press.Edelman, B. (2009). Running out of numbers: Scarcity of IP addresses and what to do about it. In S. Das, M. Ostrovsky, D. Pennock, & B. Szymanski (Eds.), Auctions, market mechanisms and their applications—First international ICST conference, AMMA 2009, Boston, MA, USA, May 8–9, 2009, Revised selected papers (pp. 95–106). Berlin, Germany: Springer-Verlag.Elmore, H., Camp, L. J., & Stephens, B. (2008). Diffusion and adoption of IPv6 in the ARIN REGION. Paper presented at the 2008 Workshop on the Economics of Information Security, June 25–28, Dartmouth College, Hanover, New Hampshire.Farinacci, D., Fuller, V., Meyer, D., & Lewis, D. (2010, April 25). Locator/ID separation protocol (LISP). (Internet Draft). Internet Engineering Task Force. Retrieved from /http://data-tracker.ietf-lisp/S.Fuller, V., Li, T., Yu, J., & Varadhan, K. (1993, September). Classless inter-domain routing (CIDR): An address assignment and aggregation strategy. (RFC 1519), Internet Engineering Task Force. Retrieved from /http://www.faqs.org/rfcs/rfc1519.htmlS.Hain, T. (2005). A pragmatic report on IPv4 address space consumption. The Internet Protocol Journal, 8(3), 2–9.Hofmann, J. (2009). Before the sky falls down: A constitutional dialogue over the depletion of Internet addresses. Paper presented at the Fourth annual symposium of the Global Internet Governance Academic Network (GigaNet), Sharm-el-Sheik, Egypt, 14 November 2009.Hubbard, K., Kosters, M., Conrad, D., Karrenberg, D., & Postel, J. (1996, November). Internet registry IP allocation guidelines. (RFC 2050), Internet Engineering Task Force. Retrieved from /http://www.faqs.org/rfcs/rfc2050.htmlS.Huitema, C. (1994, November). The H ratio for address assignment efﬁciency. (RFC 1715), Internet Engineering Task Force. Retrieved from /http://tools.ietf. org/html/rfc1715S.Huston, G. (2001). Analyzing the Internet BGP routing table. The Internet Protocol Journal, 4(1), 2–15.Huston, G. (2009). BGP in 2008. Retrieved from /http://www.potaroo.net/ispcol/2009-03/bgp2008.htmlS.Internet Architecture Board. (2001, September). IAB/IESG recommendations on IPv6 address allocations to sites. (RFC 3177), Internet Engineering Task Force. Retrieved from /http://www.ietf.org/rfc/rfc3177.txtS.Lehr, W., Vest, T., & Lear, E. (2008). Running on empty: The challenge of managing Internet addresses. Paper presented at the 36th annual telecommunications policy research conference, September 27, 2008, George Mason University, Arlington, VA, USA. Retrieved from /http://cfp. mit.edu/publications/CFP_Papers/Lehr%20Lear%20Vest%20TPRC08%20Internet%20Address%20Running%20on%20Empty.pdfS.Meng, X., Xu, Z., Zhang, B., Huston, G., Lu, S., & Zhang, L. (2005). IPv4 address allocation and the BGP routing table evolution. ACM SIGCOMM Computer Communication Review, 35(1), 71–80.Meyer, D., Zhang, L., & Fall, K. (2007, September). Report from the IAB workshop on routing and addressing. (RFC 4984), Internet Engineering Task Force. Retrieved from /http://www.ietf.org/rfc/rfc4984.txtS.Millet, S., & Huston, G. (2005, August). Proposal to amend APNIC IPv6 assignment and utilisation requirement policy. (Proposal No. Prop-031-v001). Retrieved from /http://mailman.apnic.net/mailing-lists/sig-policy/archive/2005/09/msg00013.htmlS.Mueller, M. (2002). Ruling the root: Internet governance and the taming of cyberspace. Cambridge, MA: MIT Press.Mueller, M. (2008). Scarcity in IP addresses: IPv4 address transfer markets and the regional Internet address registries. Internet Governance Project. Retrieved from /http://internetgovernance.org/pdf/IPAddress_TransferMarkets.pdfS.Mueller, M., & Kuerbis, B. (2008). Regional address registries, governance and Internet freedom. Internet Governance Project. Retrieved from /http:// internetgovernance.org/pdf/RIRs-IGP-hyderabad.pdfS.Narten, T. (2005, June). Issues related to the management of IPv6 address space. IETF Internet Draft. Retrieved from /http://ietfreport.isoc.org/all-ids/ draft-narten-iana-rir-ipv6-considerations-00.txtS.North, D. C. (1990). Institutions, institutional change, and economic performance. Cambridge, UK: Cambridge University Press.O’Dell, M. (1997). GSE—An alternate addressing architecture for IPv6, Internet Draft. Retrieved from /http://ietfreport.isoc.org/all-ids/draft-ietf-ipngwg-g seaddr-00.txtS.Ostrom, E. (2005). Understanding institutional diversity. Princeton, NJ: Princeton University Press.Perset, K. (2007). Internet address space: Economic considerations in the management of IPv4 and in the deployment of IPv6. (Report # DSTI/ICCP(2007)20/ FINAL). Paris, France: Organization for Economic Cooperation and Development.Ramadass, S. (2009). A study on the IPv6 address allocation and distribution methods. National Advanced Center for IPv6 Center of Excellence, Malaysia.Rekhter, Y., & Li, T. (1995, March). A Border Gateway Protocol 4 (BGP-4). (RFC 1771), Internet Engineering Task Force. Retrieved from /https://www.arin. net/knowledge/rfc/rfc1771.txtS.Rekhter, Y., & Li, T. (1996, October). Implications of various address allocation policies for Internet routing. (RFC 2008), Internet Engineering Task Force. Retrieved from /http://www.faqs.org/rfcs/rfc2008.htmlS.Rekhter, Y., Li, T., & Hares, S. (2006, January). A Border Gateway Protocol 4 (BGP-4). (RFC 4271), Internet Engineering Task Force. Retrieved from /http:// www.ietf.org/rfc/rfc4271S.
416 M. Mueller / Telecommunications Policy 34 (2010) 405–416Rekhter, Y., Resnick, P., & Bellovin, S. (1996). Financial incentives for route aggregation and efﬁcient utilization in the Internet. In B. Kahin, & J. H. Keller (Eds.), Coordinating the Internet (pp. 273–287). Cambridge, MA: MIT Press.Van de Velde, G., Popoviciu, C., Chown, T., Bonness, O., & Hahn, C. (2008, December). IPv6 unicast address assignment considerations. (RFC 5375), Internet Engineering Task Force. Retrieved from /http://tools.ietf.org/html/rfc5375S.Wang, Z., & Crowcroft, J. (1992, May). A two-tier address structure for the Internet: A solution to the problem of address space exhaustion. (RFC 1335), Internet Engineering Task Force. Retrieved from /http://tools.ietf.org/html/rfc1335S.