Peer-to-peer overlays: structured, unstructured, or both?
Miguel Castro, Manuel Costa, Ant Rowstron · 2004
1 Introduction There has been much interest in peer-to-peer data sharing andcontent distribution applications. They are used by millions of users and they represent a large fraction of the traffic in theInternet [29]. These applications are built on top of large-scale network overlays that provide mechanisms to discover datastored by overlay nodes. There are proposals for two types of overlays: unstructured and structured. This paper presents adetailed comparison of structured and unstructured overlays, and derives a hybrid overlay that can outperform both. Unstructured overlays, for example Gnutella [1], organizenodes into a random graph and use floods or random walks