Peer-to-Peer Economics

Introduction
Detailed modelling of the economic transactions carried out in a p2p file sharing system is, in general, a very complex task. The main reason is that participating peers should contribute different types of resources (bandwidth, storage, CPU cycles, content) with differing characteristics, the provision of which generates complex costs such as legal risks and time spent using the system. Moreover, in certain cases, there exist several intangible value generators from participating in such a system, which often involve altruism, community building, "fighting the system", and more. Actually, some of these might be part of the reason why the theoretical results of economic theory are not always compatible with the performance of real p2p applications, which seems to be acceptable even without explicit incentives for cooperation. As far as economic efficiency is concerned (i.e. maximization of social welfare), besides the complicated modelling, an additional challenging issue is the lack of information concerning the types and preferences of the individual peers, which is required for the computation of the optimal allocation of resources and cost in a p2p system. The majority of our work addresses both theoretical and implementation issues concerning the improvement of economic efficiency of a p2p file-sharing system, in respect of content availability, which we consider its most important parameter. Content availability is a public good: the copying of a file by one peer does not prevent another peer also from copying it; but contributing files to the common pool is costly. The asymptotic analysis of certain public good models for p2p file sharing suggests that when the aim is to maximize social welfare, a fixed contribution scheme in terms of the number of files shared per unity of time can be asymptotically optimal as the number of participants n grows to infinity (see [3][5][6]). Such an incentive scheme is very simple and attractive, and is also suitable for other p2p applications with similar public good charasteristics such as WLAN peering [6][7]. However, its enforcement is not straightforward in cases where no trusted software or central entity accounting for peers' transactions can be assumed and peers are free to change their identity with no cost. That is, when no sort of user memory is available to be able to identify and punish the potential free riders. A 'memory-less' p2p system should rely only on the time peers are consuming resources to ensure that they contribute adequately. BitTorrent is an example of a successful real world application focusing on bandwidth provisioning for content distribution, which implements a reciprocative incentive scheme without relying on past transactions of peers but on a direct exchange of resources (i.e. upload bandwidth). Thus, we have also studied issues related to the actual design of a p2p file sharing system, which focuses on content availability without requiring the existence of any sort of user memory or the ability to permanently expel peers from the group to enforce the efficient resource provisioning from peers [2] [4], following the implementation principle of

BitTorrent. So, we aim to improve the economic efficiency of the system without suffering from whitewashing and false trading (by proposing solutions that do not depend on long term tracking of peer behaviour). We believe time spent in the system will become a critical parameter of the contribution of peers in this context, especially as access speeds increase and people store more content in their PCs for their own use. We propose an incentive mechanism which both controls the time peers stay in the system so as not to be too short (by controlling the maximum upload bandwidth offered) and provides an incentive to offer a fixed number of files throughout that time. We then present a theoretical framework for the study of its qualitative characteristics. Using this mode, we evaluate the resulting efficiency compared to the one achieved using the theoretically optimal schemes and with one that would be achieved by an "uploads=downloads" mechanism often considered in the literature. We also provide some insights for the correct tuning of its basic parameters. See [1] (Chapter 2) for an introduction to peer-to-peer economics and a literature review.

Bibliometrics: It is ACM's intention to make the derivation of any publication statistics it generates clear to the user.

1. Publication Years = the span from the earliest year of publication on a work by this
author to the most recent year of publication of a work by this author captured within the ACM bibliographic database of computing literature (The ACM Guide to Computing Literature, also known as "the Guide". Publication Count = all works of any genre within the universe of ACM's bibliographic database of computing literature of which this person was an author. Works where the person has role as editor, advisor, chair, etc. are listed on the page but are not part of the Publication Count. Citation Count = cumulative total number of times all authored works by this author were cited by other works within ACM's bibliographic database. Almost all reference lists in articles published by ACM have been captured. References lists from other publishers are less well-represented in the database. Unresolved references are not included in the Citation Count. The Citation Count is citations TO any type of work, but the references counted are only FROM journal and proceedings articles. Reference lists from books, dissertations, and technical reports have not generally been captured in the database. (Citation Counts for individual works are displayed with the individual record listed on the Author Page.) Available for download = the total number of works by this author whose full texts may be downloaded from an ACM full-text article server. Downloads from external full-text sources linked to from within the ACM bibliographic space are not counted as 'available for download'. Downloads (6 weeks) = The cumulative number of times all works by this author have been downloaded from an ACM full-text article server over the last 6-week period for which statistics are available. The counts displayed are usually 1-2 weeks behind the current date. (6-week download counts for individual works are displayed with the individual record.) Downloads (12 months) = The cumulative number of times all works by this author have been downloaded from an ACM full-text article server over the last 12-month period for which statistics are available. The counts displayed are usually 1-2 weeks behind the

2.

3.

4.

5.

6.

7.

8. 9.

current date. (12-month download counts for individual works are displayed with the individual record.) Downloads (cumulative) = The cumulative number of times all works by this author have been downloaded from an ACM full-text article server since the downloads were first counted in May 2003. The counts displayed are updated monthly and are therefore 0-31 days behind the current date. Robotic activity is scrubbed from the download statistics. Average downloads per article = The total number of cumulative downloads divided by the number of articles (including multimedia objects) available for download from ACM's servers. Average citations per article = The total Citation Count divided by the total Publication Count.