On sampling in higher-dimensional peer-to-peer systems

Abstract

We present fully distributed algorithms for random sampling of nodes in peer-to-peer systems, extending and generalizing the work of King and Saia [Proceedings of PODC 2004] from simple Chord-like distributed hash tables to systems based on higher-dimensional hierarchical constructions, like Content Addressable Networks (CAN). We also show preliminary results on the generalization of the problem to biased sampling. In addition, we provide an extension of CAN that requires only O(1) space per node and achieves O(log n) lookup latency and message complexities. © Springer-Verlag Berlin Heidelberg 2006.