P2P Network Architectures - Peer-to-Peer Computing: Applications, Architecture, Protocols, and Challenges - page 48

the operator (i.e., the telecom company) might take a loss. An example of such
collusion is that the peers perform real uploading and downloading of data
without notifying the servers so that possibly the “levy” revenues could be
lost. Much research still has to be done to improve the design so that such
malicious actions can be detected and deterred.
3.6 Video Streaming Network Architecture
Kalogeraki, Delis, and Gunopulos [Kalogeraki et al., 2003] presented a de-
tailed qualitative and quantitative study on different architectures suitable
for serving video using a P2P environment. They made some pioneering ob-
servations, e.g., that video services would become prominent in P2P systems.
Specifically, they delineated the following guidelines for designing and imple-
menting a scalable network architecture for video streaming services.
1. The network architecture and distributed indexing mechanisms should
be designed in such a way that e cient retrieval of video data is realized.
2. Queries routing should be carefully designed so that flooding of requests
is avoided.
3. Reliability and robustness of the network have to be incorporated so that
peer dynamics (i.e., peers come and go) would not disrupt the operation
of the video services.
The first network architecture considered by Kalogeraki et al. is called
the single-index/multiple-index multiple servers (SIMS/MIMS), depicted in
Figure 3.16. The key feature of this architecture is that a number of machines
are designated as indexing servers, commonly known as trackers nowadays.
The indexing servers help participating peers to locate proper serving peers
that hold the desired video objects. Each serving peer has an admission control
manager component, deciding whether a new connection can be admitted.
Furthermore, there is also a QoS manager, taking care of the QoS negotiations
(e.g., data rate of uploading) and enforcement.
The second network architecture considered by Kalogeraki et al. is called
Multiple Independent Indexed Servers (MIIS), depicted in Figure 3.17. Here,
the major feature is that each peer keeps partial indexes for the video ob-
jects possessed by other peers. The rationale of such a design is to maintain
distributed “hooks” in the computing vicinity. Periodically, updates about lo-
cally stored new video objects are sent to connected peers so as to maintain
a consistent global view of the network.
The third network architecture considered by Kalogeraki et al. is called
the Fragmented and Multiple Servers (FAMS), depicted in Figure 3.18. This