One important task of any peer-to-peer streaming system (p2p-ss)is how to choose which peers should connect to which peers. Howwell a p2p-ss perform this task greatly influences its performance.This thesis explores how different peer selection algorithmsaffect the performance of such systems.
A framework for doing the comparisons of peer selection algorithmsis built on top of the network simulator ns2, making it possible tolater extend the simulations with new peer selection algorithms,congestion control algorithms, wireless networks, cross traffic and other. However, ns2 is a low-level simulator, hence limiting thenumber of peers in the simulations, because CPU-resources are limited.The simulations are limited to single-layered streams.
We find that a centralized selection method, which utilizes knowledgeof bandwidth capacities and routing in the network, greatlyoutperforms both simple random selection of peers, and selection ofclose peers. Even though centralized selection does not scale well,and is therefore only applicable for a limited number of peers, thisshows there is much room for improvement over basic strategies.