Incentives - Peer-to-Peer Computing: Applications, Architecture, Protocols, and Challenges - page 78

Users usually perform file selection (and hence, peer selection) with the help
of some directory system which may or may not be fully distributed. For
example, in Napster [Napster, 2009], the directory is centralized.
Using such a sharing model, the most obvious form of free-riding behavior
is that a selfish user just keeps on retrieving files from others but refuses to
share its collections (and thus, no need to expend any outbound bandwidth
for file uploading). Interestingly enough, in an empirical study using the Maze
file sharing system [Maze, 2006] performed by Yang et al. [Yang et al., 2005],
it is found that the more direct indicator of free-riding behaviors is the online
time of a user. Specifically, the online time of a selfish user in a P2P file sharing
network is on average only one-third of that of a cooperative peer.
In this section, we first briefly overview a contemporary file sharing sys-
tem called BitTorrent [Cohen, 2003]. We then survey techniques suggested for
various other P2P file sharing networks.

5.2.1.1 BitTorrent

BitTorrent [Cohen, 2003] is by far one of the most successful P2P file shar-
ing systems. A key feature in BitTorrent is that each shared file is divided
into pieces (of size 256KB each), which are usually stored in multiple differ-
ent peers. Thus, for any peer in need of a shared file, parallel downloading
can take place in that the requesting peer can use multiple TCP connections
to obtain different pieces of the file from several distinct peers. This feature
is highly effective because the uploading burden is shared among multiple
peers and the network can scale to a large size. Closely related to this par-
allel downloading mechanism is the incentive component used in BitTorrent.
Specifically, each uploading peer selects up to four requesting peers in making
uploading connections. The selection priority is based on descending order
of downloading rates from the requesting peers. That is, the uploading peer
selects four requesting peers that have the highest downloading rates. Here,
downloading rate refers to the data rate that is used by a requesting peer in
sending out pieces of some other file. Thus, the rationale of this scheme is to
provide incentive for each participating peer to increase the data rate used
in sending out file data (i.e., uploading, or, in BitTorrent's term, unchoking).
There are other related mechanisms (e.g., optimistic unchoking), which are
described in detail in [Cohen, 2003, Qiu and Srikant, 2004].
Qiu and Srikant [Qiu and Srikant, 2004] performed an indepth analysis of
BitTorrent's incentive mechanism. By using an intricate and accurate model,
it is shown that a Nash equilibrium exists in the upload/download game in
BitTorrent. At the equilibrium, each peer sets its uploading data rate to be
its physical maximum uploading rate (i.e., each peer is fully cooperative).
On the other hand, due to the usage of the optimistic unchoking mechanism
(a fifth requesting peer is randomly selected in the uploading process, for
details, see [Cohen, 2003, Qiu and Srikant, 2004]), a free-rider can potentially
achieve 20% of the possible maximum downloading rate. This theoretical result