Incentives - Peer-to-Peer Computing: Applications, Architecture, Protocols, and Challenges - page 96

TABLE 5.2: A qualitative comparison of different incentive approaches for
P2P file sharing.

Payment Auction Exchange Reciprocity Reputation

Example Hauscheer Gupta and Anagnostakis BitTorrent Sun and

et al.
[Hausheer
et al., 2003]

Somani
[Gupta and
Somani,
2004]

and Green-
wald [Anag-
nostakis and
Greenwald,
2004]

Garcia-
Molina [Sun
and Garcia-
Molina, 2004]

Practical Complex Complex Easy Easy Easy

Implementation
Security High High Low Low Medium

Requirement
Centralized

Yes Yes No No Yes/No

Authority
Required
Scalability Medium Low High High High

Of course, each supplying peer may also be a recipient of another logically
different streaming session. With this streaming model, each requesting peer
needs to actively contact several supplying peers in order to start a session.
This streaming model is based on a practical observation that peer re-
quests are asynchronous and each peer's communication capability is different
(i.e., heterogeneous streaming capabilities), as illustrated in Figure 5.11 where
peers with different communication supports (e.g., DSL, dial-up, etc.) initiate
streaming requests at different times.
A layer-encoded media streaming process is assumed, as shown in Fig-
ure 5.12. As can be seen, each peer performs buffered playback so that the
received media data are kept in a buffer and thus, can be used for streaming
to other later-coming peers. For example, H 1 initiates a streaming session first
and thus, it is served solely by the server. Peer H 2 starts its session next and
so it can request H 1 , which has buffered some media data, together with the
server to send it the required data. Similarly, H 3 can stream from H 1 and
H 2 without the server as it starts its session just-in-time to use the buffered
data from the two earlier peers. On the other hand, H 4 , which starts too late,
cannot stream from H 1 and H 2 . Instead, it receives media data from H 3 and
the server.
Each potential supplying peer has only a limited capacity and thus, an
admission control mechanism is needed. Peers in the system are classified into
N classes according to the different levels of uploading bandwidth available at
the peers. Each potential supplying peer P S maintains an admission probabil-
ity vector: (P r[1], P r[2], . . . , P r[N ]). Here, a smaller index represents a class
with a larger uploading bandwidth. Suppose P S is itself a class-k peer. Then,
its probability vector is initialized as follows:

•For 1≤i≤k, P r[i] = 1.0;