=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= This message was forwarded through the Red Rock Eater News Service (RRE). You are welcome to send the message along to others but please do not use the "redirect" option. For information about RRE, including instructions for (un)subscribing, see http://dlis.gseis.ucla.edu/people/pagre/rre.html =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= Peer-to-Peer and the Promise of Internet Equality Philip E. Agre Department of Information Studies University of California, Los Angeles Los Angeles, California 90095-1520 USA pagreat_private http://dlis.gseis.ucla.edu/pagre/ This is a draft. You are welcome to forward it, but please do not quote from it. Comments appreciated. Version of 15 November 2001. 2500 words. Technologies often come wrapped in stories about politics. In the case of peer-to-peer technologies on the Internet, the standard story goes like this: Once the world was centralized under the control of a top-down hierarchy. Then came the Internet, a decentralized network architecture that lets everyone build their own services and prevents anyone from regulating them. Peer-to-peer (P2P) technologies deliver on the Internet's promise, and they will lead us to a decentralized world of freedom and equality. I propose to analyze this story. An initial problem is the term "peer-to-peer". As the story suggests, it can be hard to distinguish P2P computing from the Internet in general. For example, why isn't e-mail peer-to-peer? And if SETI@Home (Anderson 2001) is an example of P2P computing despite its centralized structure (a central server that interacts with numerous personal computers that contribute their spare cycles), then why isn't the Web also P2P, given that its client-server structure is much less centralized? Shirky (2001) proposes that P2P systems form a distinct category because they use mechanisms other than the domain name system (DNS) to identify and track their participants. Yet this definition seems arbitrary. Whatever the ills of the DNS, surely there exist other potential tracking mechanisms that are even worse. In reality, I suggest, P2P does not name a distinct category of distributed computing systems. Rather, it signifies a certain project: pushing the design of all computing systems toward a fully decentralized ideal. Yet much remains unclear. To begin with, observe that the standard story shifts between two kinds of "center", architectural and institutional. By "architecture" I mean the concepts that are designed into a technology. Examples include the von Neumann serial processor, the client-server distinction, and the relational database. By "institution" I mean the concepts that organize language, rules, job titles, and other social categories in a given sector of society. Examples include the concepts of "patient", "case", and "disease" in the medical system. Architectures and institutions are often related, and systems analysts have developed sophisticated methods of translating institutional concepts into system architectures. The standard story suggests, very simply, that decentralized architectures will bring about decentralized institutions (see, e.g., Gilder 1992). Yet this hardly follows. Architectures and institutions are often shaped to fit another another, but they are still different sorts of things. As a means of evaluating the prospects for P2P, therefore, I will briefly present four theories of the relation between architectures and institutions. Each theory will be associated with a particular theorist of institutions. 1. Veblen Thorstein Veblen wrote during the Progressive Era, when tremendous numbers of professional societies were being founded, and he foresaw a society that was organized rationally by engineers rather than through the speculative chaos of the market. Veblen was impressed by a profession's ability to pool knowledge among its members, and he emphasized the collective learning process through which industry grows. (On Veblen's theory, see Hodgson (1999).) Veblen's theory resembles some of the claims made for the Internet: open information, universally shared, giving rise to a group mind. In fact, the rise of professions was facilitated by the communications and transportation infrastructures of the 19th century. As Chandler (1977) observes, the first modern professional associations were organized by railroad employees, who used the new telegraph and the railroad infrastructures, as well as printed newsletters and organized conferences, to build new institutions of knowledge-sharing. The infrastructures, in turn, had gone through a tumultuous history of decentralized innovation and increasing centralization under the control of large firms. 2. Hayek Friedrich Hayek was an Austrian economist who withdrew from technical research to provide intellectual ammunition for the fight against communism. His most famous argument is that no centralized authority could possibly synthesize all of the knowledge that the participants in a complex market use in making their allocation decisions (Hayek 1963). This emphasis on knowledge was outside the mainstream of economic thought at the time, and it remains largely so even today. But Hayek was not an anarchist. He argued that a market society requires an institutional substrate that upholds principles such as the rule of law (Hayek 1960). A productive tension is evident in Hayek's work: he is attracted to notions of self-organization that seem like the opposite of governmental control, but he is also aware that self-organization presupposes institutions generally and government institutions in particular. Hayek's work, like Veblen's, challenges us to understand what a "center" is. In some cases, intuitions are clear. French society is highly centralized. Switzerland has been remarkably decentralized for centuries. And the federal systems of the United States and Germany lie somewhere in the middle. In each case, we reckon degrees of centralization by the constitutional distribution of political authority. But centers and centralization can be understood in other ways. Observe that institutions, like architectures, are typically organized in layers. Legislatures and courts are institutions that create other institutions, namely laws. Contract law is an institution, but then so are individual contracts. Internet protocols, likewise, are organized in layers, each of which creates ground rules for the ones above it. Do the more basic layers count as "centers"? Yes, if they must be administered by a centralized authority. Yes, if global coordination is required to change them. No, if they arise in a locality and propagate throughout the population. At least sometimes, then, centralization on one layer is a precondition for decentralization on the layers above it. Complex markets systems, for example, need their underlying infrastructures and institutions to be coordinated and standardized. Yet this kind of uniformity has generally been imposed by powerful governments and monopolies. The conditions under which decentralized systems can emerge, therefore, are complicated. Consider the case of the Internet. Despite its reputation as the very model of decentralization, the institutions and architecture of the Internet nonetheless have many centralized aspects, including the DNS, the IETF, and Microsoft's control over the desktop software market. But let us consider one aspect of the Internet in particular: the end-to-end principle (Saltzer, Reed, and Clark 1984), which moves complexity out of the network itself and into the hosts that use it. In one sense, this principle is nothing but layering. Each layer in the Internet protocol stack is kept simple, and new functionalities are assigned to newly created layers atop the old ones. In another sense, however, the end-to-end principle shifts complexity away from the centralized expertise of network engineers, placing it instead on the desktops of end-users -- the very people who are least able to manage it. Much of the Internet's history, consequently, has consisted of attempts to reshuffle this complexity, moving it away from end-users and into service providers, Web servers, network administrators, and so on. As this history makes clear, one layer of an architecture can have different properties from the layers above and below it. A decentralized network can support centralized services, or vice versa. Thus, for example, the asymmetrical client-server architecture of the Web sits atop the symmetrical architecture of the Internet. The peer-to-peer movement promises to move the entire network toward a decentralized ideal. In doing so, it must confront various types of centralization that are inherent in certain applications. For example, if users contend for access to the same physical resource, some kind of global lock will be needed. Most markets have this property. Some mechanisms do exist for sharing a scarce resource without an architecturally centralized lock; examples include the backoff algorithms that both Ethernet and TCP use to control network congestion. Research on distributed services has long sought to replicate documents while avoiding the danger of divergent changes (e.g., Dewan 1999). So the obvious solution of complete architectural centralization is hardly the only option. Even so, it is a profound question how thoroughly the functionality of a market mechanism like NASDAQ, eBay, or SABRE can be distributed to buyer/seller peers. It is also unclear how useful such a radical decentralization would be to the market participants. 3. North For Douglass North (1990: 3), an institution can be understood by analogy to the rules of a game. The rules of baseball, for example, define such categories as a "pitcher", "strike", and "infield fly". An institution, in this sense, is categorical structure that allows people to coordinate their activities. The institution defines social roles and creates a terrain upon which individuals navigate. In particular, the institution creates incentives, such as the profit motive, that tend to channel participants' actions. Taking markets as his main example, North suggests that the rules of the game change only slowly and incrementally. (For an application of North's theory to the DNS controversy, see Mueller (2000).) As an example, consider the institutional context in which the ARPANET and Internet arose (Abbate 1999). In their attempt to create a decentralized computer network, the program managers at ARPA had an important advantage: they controlled the finances for a substantial research community. ARPA made the rules, and they consciously created incentives that would promote their goals. They compelled their contractors to use the ARPANET (1999: 46, 50, 55), and they drove the adoption of electronic mail by methods such as being accessible to their contractors only through that medium (1999: 107-110). Later on, they funded implementation of TCP/IP on many vendors' machines (1999: 143) and imposed a death march on their contractors for the transition to TCP (1999: 140-142). This centralized institutional environment had subtle consequences for the decentralized architecture it produced. Because of ARPA's authority, everyone took for granted that the ARPANET's user community was self-regulating. This feature of the institution is reflected in the poor security of the Internet's electronic mail standards. When the Internet became a public network, the old assumptions no longer applied. Lasting security problems were the result. Another example is found in Orlikowski's (1993) celebrated study of Lotus Notes in a large consulting firm. Notes may not be a wholly peer-to-peer architecture, but its success was due largely to its replication strategy, which is crucial for distributed document- sharing. The CIO of this firm assumed that he could ensure widespread adoption simply by making the software available on employees' computers. That did not happen. Orlikowski identified two problems: (1) most employees were familiar with traditional tools such as electronic mail, so they used Notes only for those familiar purposes, and (2) most of the consultants were promoted on the basis of the distinctive practices that they had built as individuals, so they had few incentives to share their knowledge. Only when the company began evaluating employees on their use of Notes, therefore, did adoption become widespread. Once again, centralized authority was required to institutionalize decentralized knowledge-sharing. 4. Commons John Commons was a Progressive Era economist who eventually trained many of the leaders of the New Deal. Guided by his union background and the democratic ideals of his time, Commons (1934) viewed every social institution as a set of working rules defined by collective bargaining. After all, every institution defines a set of social roles (doctor-patient, teacher-student, landlord-tenant, and so on), and each social role defines a community (for example, the community of doctors and the community of patients). Commons (1924) argues that each group develops its own culture and practices, which eventually become codified in law. Commons' theory helps to explain the development of technical architectures. Consider, for example, the classic institutional analysis by Danziger, Dutton, Kling, and Kraemer (1982) of the development of computer systems in American local governments. These authors observed that computer architectures are rarely neutral. Whose functionalities should the system support? Who should gain information about whom? The design process, therefore, is inevitably political. Based on survey and interview studies, the authors asked which factors determined the interests a new computer system would end up serving. They did find that every affected group had some input to the decision-making process. But they concluded that new systems ended up serving the interests of whichever group (for example, the mayor, the financial department, or the police) already held power. Commons would view this situation as pathological. He disagreed with Marx's vision of history as the inevitable victory of one social class over all others, and preferred a vision of collective bargaining among evenly matched groups. For this reason, he might have found more hope in the current war over music distribution. The unexpected growth of Napster set off an institutional revolution in the music industry, and Napster's subsequent decline under legal attack should provoke reflection about that revolution's nature. Napster had a fatal flaw: although it provided a viable architecture for music sharing, it did not provide a viable institution for allowing musicians to make a living. Some bands can make money from live performance or merchandise, but most bands -- if they make money at all -- still rely on record sales. The collective bargaining process that Napster has set in motion, therefore, has at least three parties: musicians, fans, and record companies. As in every negotiation, each party has its own political problems -- comprehending the situation, getting organized, adopting a common position, coordinating its actions, delegating authority to a trusted representative, and so on. The negotiation takes the form of an "ecology of games" (Dutton 1992): conflicts in many venues, including legislatures, courts, and standards organizations. What is needed, clearly, is an alternative institutional model that connects musicians and fans in new ways, ideally without the market dysfunctions that lead to abusive record industry contracts. This alternative institution for music distribution will presumably depend on new technical architectures. Yet, for the moment, most technical development is aimed at protecting a Napster-like model from the legal assaults of the record companies. It remains to be seen whether a thoroughly peer-to-peer "sharing" architecture can avoid being shut down, particularly if monopolies such as Microsoft change their own architectures to suit the record companies' needs. A more important question, though, is whether the drive toward fully decentralized "sharing" bears any useful relationship to the real problem of connecting musicans and fans in an economically viable way. What has been learned? Decentralized institutions do not imply decentralized architectures, or vice versa. Indeed, the opposite is just as arguably the case. The drive toward decentralized architectures need not serve the political purpose of decentralizing society, and it can even be destructive. Architectures and institutions will inevitably coevolve, and to the extent they can be designed, they should be designed together. The peer-to-peer movement understands that architecture is politics, but it too often assumes that architecture is a substitute for politics. Radically improved information and communication technologies do open new possibilities for institutional change. To explore those possibilities, though, technologists will need better ideas about institutions. References Janet Abbate, Inventing the Internet, Cambridge: MIT Press, 1999. David Anderson, SETI@home, in Andy Oram, ed, Peer-to-Peer: Harnessing the Power of Disruptive Technologies, O'Reilly, 2001. Alfred D. Chandler, Jr., The Visible Hand: The Managerial Revolution in American Business, Cambridge: Harvard University Press, 1977. John R. Commons, Legal Foundations of Capitalism, New York: Macmillan, 1924. John R. Commons, Institutional Economics: Its Place in Political Economy, Madison: University of Wisconsin Press, 1934. James N. Danziger, William H. Dutton, Rob Kling, and Kenneth L. Kraemer, Computers and Politics: High Technology in American Local Governments, New York: Columbia University Press, 1982. Prasun Dewan, Architectures for collaborative applications, in Michel Beaudouin-Lafon, ed, Computer Supported Co-Operative Work, Chichester, UK: Wiley, 1999. William H. Dutton, The ecology of games shaping communications policy, Communication Theory 2(4), 1992, pages 303-328. George Gilder, Life after Television, New York: Norton, 1992. Friedrich A. Hayek, The Constitution of Liberty, Chicago: University of Chicago Press, 1960. Friedrich A. Hayek, Individualism and Economic Order, Chicago: University of Chicago Press, 1963. Geoffrey M. Hodgson, Economics and Utopia: Why the Learning Economy Is Not the End of History, London: Routledge, 1999. Milton Mueller, Technology and institutional innovation: Internet domain names, International Journal of Communications Law and Policy 5(1), 2000. Available on the Web at <http://www.IJCLP.org/5_2000/ijclp_webdoc_1_5_2000.html>. Douglass C. North, Institutions, Institutional Change, and Economic Performance, Cambridge: Cambridge University Press, 1990. Wanda J. Orlikowski, Learning from Notes: Organizational issues in groupware implementation, The Information Society 9(3), 1993, pages 237-250. Jerome W. Saltzer, David P. Reed, and David D. Clark, End-to-end arguments in system design, ACM Transactions in Computer Systems 2(4), 1984, pages 277-288. Clay Shirky, Listening to Napster, in Andy Oram, ed, Peer-to-Peer: Harnessing the Power of Disruptive Technologies, O'Reilly, 2001. end
This archive was generated by hypermail 2b30 : Thu Nov 15 2001 - 11:48:43 PST