Incentives - Peer-to-Peer Computing: Applications, Architecture, Protocols, and Challenges - page 83

experiences in that a satisfactory transaction leads to an increase in the reserve
prices while an unsatisfactory one leads to a decrease. From a seller's point
of view, these prices are also exponentially decreased as time goes by. Under
this model, simulation results indicate that a free-rider will quickly deplete its
budget. On the other hand, the price of a direct answer is found to be much
higher than that of a referral.
Courcoubetis and Weber [Courcoubetis and Weber, 2006] recently reported
an indepth analysis of the cost in sharing in a P2P system. In their study, a
P2P sharing system is modeled as a community with an excludable public
good. Furthermore, the public good is assumed to be nonrivalrous, meaning
that a user's consumption of the public good does not decrease the value
of the good. Such a model is suitable for a P2P file sharing network, where
the excludable public good is the availability of shared files. The model is
also considered as suitable for a P2P wireless LAN environment, where the
excludable public good is the common wireless channel. With their detailed
modeling and analysis, an important conclusion is derived: each peer only
needs to pay a fixed contribution, in terms of service provisioning (e.g., a
certain fixed number of distinct files to be shared by other peers), in order
to make the system viable. Such a fixed contribution is to be computed by
some external administrative authority (called a “social planner”) by using
the statistical distribution of the peers' valuations of the public good.

5.2.1.5 Reciprocity and Reputation-Based Systems

Feldman et al. [Feldman et al., 2004a] suggested an integrated incen-
tive mechanism for effectively deterring (or penalizing) free-riders using a
reciprocity-based approach. Specifically, the proposed integrated mechanism
has three core components: discriminating server selection, maxflow-based
subjective reputation computation, and adaptive stranger policies.
In the discriminating server selection component, each peer is assumed
to have a private history of transactions with other peers. Thus, when a file
sharing request is initiated, the peer can select a server (i.e., a file owner)
from the private history. However, in any practical P2P sharing network, we
can expect a high turnover rate of participation. That is, a peer may only
be present in the system for a short time. Thus, when a request needs to be
served, such a departed peer would not be able to help if it is selected. To
mitigate this problem, a shared history is to be implemented. That is, each
peer is able to select a server from a list of global transactions (i.e., not just
restricted to those involving the current requesting peer). A practical method
of implementing shared history is to use a distributed hash table-(DHT-) based
overlay networking storage system [Stoica et al., 2001b]. Specifically, a DHT
is an effective data structure to support fast look-up of data locations.
A problem in turn induced by the shared history facility is that collusion
among non-cooperative users may take place. Specifically, the non-cooperative
users may give each other a high reputation value (e.g., possibly by report-