Information Technology Reference
In-Depth Information
centive system called Selfish Link-based InCentive (SLIC), which is based on
pairwise reputation values. Specifically, any peer u maintains a reputation
value W (u, v) for each of its neighbor peer v, where the reputation value is
normalized such that 0≤W (u, v)≤1. Here, “neighbor” means a peer v
currently having a logical connection with u and thus, such a peer v can po-
tentially request for service from u. With these reputation values, the peer u
can then allocate the uploading bandwidth to any requesting neighbor peer v
with a value of W (u, v)/
i W (u, i). The reputation value W (u, v) is updated
periodically based on an exponential averaging method.
Under this model, Sun and Garcia-Molina [Sun and Garcia-Molina, 2004]
observed that each peer has the incentive to do some or all of the following, in
order to increase its reputation values as perceived by other peers (and hence,
enjoy a better quality of service).
•Sharing out more file data;
•Connecting to more peers (to increase the opportunities for serving oth-
ers);
•Increasing its total uploading capacity.
5.2.1.6
Penalty-Based Approaches
Feldman et al. [Feldman et al., 2004b] also investigated disincentive mech-
anisms that can discourage free-riding. Specifically, they considered various
possible penalty schemes in deterring free-riders. A simple model is used. At
the core of the model, each user i in the P2P sharing network is characterized
by a positive real-valued type variable, denoted as t i . Another key feature of
the model is that the cost of contributing is equal to the reciprocal of the cur-
rent percentage of contributors, which is denoted as x. Thus, for any rational
user with type t i , the user will choose to contribute if 1/x < t i and free-ride
if 1/x≥t i .
Furthermore, the benefit each user derived from the P2P network is as-
sumed to be of the form αx β , where β≤1 and α > 0. With this benefit func-
tion, the system performance is defined as the difference between the average
benefit and the average contribution cost. Specifically, system performance is
equal to: αx β −1.
Even with the simplistic model described above, Feldman et al. provided
several interesting conclusions. Firstly, it is found that excluding low type
users can improve system performance only if the average type is low and α
is large enough. Unfortunately, exclusion is impractical because a user's type
is private and thus, cannot be determined accurately by other peers. It is
then assumed that free-riding behaviors are observable (i.e., free-riders can be
identified). Such free-riders are then subject to a reduction in quality of service.
Quantitatively, the benefit received by a free-rider is reduced by a factor of
(1−p), where 0 < p≤1. A simple implementation of this penalty is to exclude
a free-rider with a probability of p. The second interesting conclusion is that
Search WWH ::




Custom Search