P2P Network Architectures - Peer-to-Peer Computing: Applications, Architecture, Protocols, and Challenges - page 29

work needs efforts to maintain its “shape.” If the overhead of maintain-
ing the network's characteristics (e.g., diameter, etc.) is low, then service
disruptions would be reduced to a minimum.
QoS provisioning. Combining the above two nice features, we could possi-
bly go one step further—to provide certain guarantees of QoS to peers.
For instance, it would be nice if we can guarantee a certain playback
delay for every new P2P video streaming peer.
Autonomy. By its nature, each peer in the network is autonomous, in the
sense that it is under the user's full control. Consequently, if the network
architecture allows for such autonomy to the greatest extent, then more
users would like to participate.
Robustness. The network should also be resilient to changes, e.g., peer dy-
namics, peer failures, network failures, etc.
Yet it is clear that the above nice features are conflicting, if not totally mu-
tually exclusive. For example, allowing for maximum autonomy is in conflict
with QoS provisioning, in most cases. Furthermore, considering a structure
network such as a DHT, it is mandatory for each peer to adhere to some pro-
tocol rules (e.g., responsible for handling a pre-defined range of data items)
in order to realize the nice feature of provably good performance. Thus, au-
tonomy is sacrificed in this case. On the other hand, a totally unstructured
network allows for maximum autonomy. Yet such a system can hardly pro-
vide any QoS guarantee or any provably good performance. Consequently, a
“holy grail” research problem is to come up with low-overhead mechanisms
to provide QoS guarantees in an unstructured autonomous P2P network ar-
chitecture.
Furthermore, there are several other nice features that we must consider
if we consider the existence (inevitable) of malicious users. These features are
incentives, trust, and security. Incentives are important to encourage coop-
erative and constructive, instead of malicious, behaviors. Trust is necessary
among peers in order to accept the data in exchange. Proper security mea-
sures are needed if there are really some malicious users participating in the
network, possibly trying to undermine the normal operations of the system.
These are the topics of later chapters in this topic.
In the following we provide a brief survey of different network architecture
designs.
3.2 Structured P2P Systems
In this section, we explain why structured network architecture came about
and describe several representative approaches.