Topology Control - Peer-to-Peer Computing: Applications, Architecture, Protocols, and Challenges - page 57

We then describe an approach which is based on inference actions per-
formed by each peer in a specialized network-coding-based P2P system. Net-
work coding is a bandwidth e cient packet transmission technique and it
allows peers to deduce network bottlenecks.
Finally we also briefly describe a recently proposed topology control tech-
nique designed for a wireless P2P system, for which energy e ciency is a prime
concern.

4.2 A General Framework for Distributed Topology

Control
Singh and Haahr [Singh and Haahr, 2006] suggested using Schelling's
model [Schelling, 1971] to dynamically adjust the topology of a P2P network.
Specifically, Schelling (an economist) observed that the existence of segregated
neighborhoods in the U.S. was neither caused by a central authority nor by
the desire of people to stay away from dissimilar people. Instead, the segre-
gation is the cumulative effect of simple actions by individuals who want at
least a certain proportion of their neighbors to be similar to themselves.
In Schelling's abstract model [Singh and Haahr, 2006], the world is mod-
eled as a grid. Approximately two-thirds of the cells in the grid are populated
by blue or red turtles. The remaining cells are empty. Each cell can host a
maximum of one turtle. In the beginning, a random number of blue and red
turtles are randomly distributed on the grid. All the turtles desire at least
a certain percentage of their neighbors to be of the same color. If a turtle is
dissatisfied with its neighbors, it moves to an adjacent empty cell (if available)
chosen randomly. This process repeats until all the turtles are satisfied with
their neighbors. The resulting segregation is an emergent behavior caused by
the desire of the turtles to ensure a certain minimum percentage of their neigh-
bors are the same color as themselves. Schelling's model is thus applicable in a
P2P network because each peer lacks a global picture of the network topology.
Furthermore, in the model, grouping is maintained even when turtles join or
leave the system, which makes it attractive for the dynamic environments of
P2P networks.
A similar approach has also recently been suggested by Hariri et al. [Hariri
et al., 2007] for application level network overlay topology control in a mas-
sively multiplayer online game (MMOG). There are many other topology con-
trol schemes, such as Auvienen et al. [Auvinen et al., 2007], that also fit in
this general framework.