Security Issues - Peer-to-Peer Computing: Applications, Architecture, Protocols, and Challenges - page 162

As illustrated by Kumar et al.'s numerical results, the above solution gen-
erates very accurate data as compared to a simulated system, especially when
M is large (e.g., 100,000). Using this model, many interesting conclusions can
be drawn. For instance, the number of corrupted copies reaches a peak and
then quickly drops off to zero in an exponential manner. Similar mathemati-
cal analysis is also given by Kumar et al. [Kumar et al., 2006] for the Version
Centric Downloading model.
Yang et al. [Yang et al., 2008] also proposed a content pollution dynamics
model for live video streaming systems. They found that the most critical
factors are the access bandwidth and the degree of participating peers, but
not the number of initial polluters.
To combat pollution attacks, a typical approach is still based on reputation
scores associated with the files as well as the peers that are sharing them.
Specifically, when a peer selects a certain file for downloading, it first requests
for a weighted sum of votes for the file from a set of peers. The latter sends
the requesting peer a vote (e.g., +1 or−1) so that a weighted sum can be
computed. The weights used represent (or are derived from) the reputations
of these responding peers. The systems Credence and Scrubber as described
in [Costa and Almeida, 2007] are based on such ideas. A major performance
criterion for these systems is the convergence rate. As shown in the simulation
results in [Costa and Almeida, 2007], it typically takes several days for the
systems to converge to accurately identify true copies from polluted copies.
7.3 Buffer Map Cheating
Li et al. [Li et al., 2009] studied the problem of buffer map cheating in
P2P video streaming systems. Specifically, they considered the situation where
some selfish peers lie about their buffer map contents in that some available
chunks are held back. The rationale of this selfish behavior is that the upload-
ing burden can be reduced. However, the streaming quality of the requesting
peers could also be reduced.
To combat such malicious behaviors, Li et al. [Li et al., 2009] proposed
a simple incentive scheme, which works by sorting the requests at a chunk
uploading peer in descending order of previous contribution levels. Conse-
quently, if a selfish peer holds back some chunks, its contribution level would
be reduced, and thus, it will be placed at a later position in the request queue
when it requests chunks from another peer. Li et al.'s simulation results show
that the proposed simple incentive scheme works quite well in deterring selfish
behaviors.