In our ongoing work, we are characterize the nature of peer dynamics
in peer-to-peer systems. While some previous studies have touched on
this topic while studying other properties of peer-to-peer systems, to
our knowledge this is the first study to focus entirely on churn. Our
goal in this study is to form accurate and detailed characterizations
which can be used for simulation and analysis.
Using data gathered from Gnutella, Kad, and BitTorrent, we are making
a comparative study to locate similarities and differences of user
behavior across these systems. Our initial results show that several
dynamic properties exhibit similarities across all three systems,
suggesting that these properties are user-driven rather than
application-specific.