Creating an adaptive network of hubs using Schelling's model

Reviewer: Andrew Robert Huber

Peer-to-peer (P2P) systems organize peers into virtual networks based on the underlying physical Internet. These self-organized networks may result in inefficient network topologies for common P2P tasks such as searching and clustering. To solve this problem, this article describes an abstract algorithm based on Schelling’s model. This abstract algorithm is used to create a family of algorithms that, when executed by the peers, create an efficient virtual network of hubs by dynamically adapting the original P2P network. In the abstract algorithm, each peer periodically calculates whether it is satisfied with its local view of the overlay network topology (for example, its neighbor count). If not, the peer takes steps to adapt the topology, for example by adding or removing neighbors. For the particular example algorithm in the article, hubs (high-bandwidth peers connected to other hubs) are satisfied if connected to more than zero but no more than a maximum number of other hubs and no more than a maximum number of total peers. Hubs will drop nonhub neighbors and add hubs as neighbors until all hubs are satisfied. Simulations on networks with 100 to 1,000 peers with a maximum number of neighbor hubs from one to ten are reported on, showing results on how the networks converge, the numbers of messages passed to reach convergence, and other properties of interest. This is a concise, well-written article with interesting results that should be relevant to practitioners and researchers working with networks or P2P systems. Online Computing Reviews Service