A Reciprocation-Based Economy for Multiple Services in Peer-to-Peer Grids Miranda Mowbray † , Francisco Brasileiro ‡ , Nazareno Andrade ‡ , Jaindson Santana ‡ , Walfredo Cirne ‡ † Hewlett-Packard Laboratories Bristol ‡ Universidade Federal de Campina Grande Filton Rd, Stoke Gifford Av. Apr´ ıgio Veloso, s/n - 58.109-970 Bristol BS34 8QZ, UK Campina Grande, Brazil miranda.mowbray@hp.com {fubica,nazareno,jaindson,walfredo}@dsc.ufcg.edu.br Abstract In this paper we study reciprocation-based mechanisms to encourage donation in peer-to-peer grids in which mul- tiple services, such as processing power and data transfers, are shared explicitly. We have modeled such a system and established how peers should assess whether it is profitable to exchange services with another peer, an issue that is not present in the single service case. Unfortunately, this as- sessment relies on information provided by untrustworthy peers. As an alternative, we have extended, to the case of multiple services, a reciprocation-based mechanism which uses only reliable information gathered locally. We have assessed this mechanism by simulating scenarios in which services are exchanged that are combinations of two differ- ent basic services. In the explored scenarios the mechanism performs very well, and can marginalize free riders even when the cost to peers of donating a service is nearly as large as the utility gained by receiving it. 1. Introduction A computational grid is a federation of sites across dif- ferent administrative domains which shares computational services. A peer-to-peer grid is a large-scale, free-to-join computational grid in which participants do not necessar- ily know or trust each other. In a peer-to-peer grid, each peer represents a site; whenever the site has idle capacity it provides services to other peers, and whenever the site needs more computational services, it requests them from all other peers. Examples of projects building such grids are OurGrid [9, 17], Cluster Computing On the Fly [16] and the Self-organizing Flock of Condors proposed by Butt et al. [7]. Peers gain utility from using the services of other peers, and there is a cost for providing a service to the grid. This cost is incurred to maintain the hardware and software needed by a site to provide services to the grid, and through the security risk posed to this site when providing such ser- vices to non-trusted parties. These costs must be low in comparison with the benefit obtained by joining the grid, and they can be lowered by reducing the effort needed to install and maintain the grid middleware, as well as by im- provements to security mechanisms. However, it is naive to assume that all costs can be eliminated. In contrast to most grids currently in production, a peer- to-peer grid cannot rely on off-line negotiations or trust chains comprising all participants in the grid to enforce co- operative behavior. In this setting, an incentive mechanism plays a key role in promoting service provision to the grid. If there is a non-zero cost for donating services and a peer can obtain the same amount of service no matter how much it serves other peers, then peers have an economic incentive to contribute nothing and free ride. This behavior is indeed found in peer-to-peer file sharing systems [2, 14]. Free rid- ing reduces the amount of service available in the grid, and diminishes the utility of the system for users of resource- intensive applications; most users of computational grids run such applications. This motivates the use of an incentive mechanism to promote collaboration in peer-to-peer grids. In previous work, we proposed the Network of Favors, a reciprocation-based mechanism for peer-to-peer grids in which a single service is shared [3, 4]. In the Network of Favors, peers exchange donations of services. Peers always donate their spare service, and decide whom to serve based solely on the record of their past bilateral interactions with peers requesting the service. We have shown that in a peer-to-peer grid sharing a sin- gle service – access to processing power – this autonomous behavior provides an incentive for peers to contribute as much as possible [3, 4]. This happens because peers who contribute more get more in return when they make re- quests. Since the balance of past interactions is very little information, a peer can keep track of its interactions with a