A peer-to-peer object location system is an evolving set of computers cooperating to store objects.
A reasonable system should easily adapt when computers join or leave the network (self-organization), reliably find objects (completeness), and ensure that no computer works too hard (load balance).
Searches in this network should find nearby copies of objects when
possible: a searcher in Berkeley looking for
an object on the Berkeley subnetwork should find the
object without ever sending a message outside of Berkeley.
In this thesis, we describe the first techniques to maintain these
properties. Our performance depends on an upper bound on the growth
rate of the network, which can be high in some cases. We further
build an adaptive scheme that depends only on a local version
of the growth rate. As a result, the bad areas of the network do not
force high resource usage everywhere. We also describe techniques to
make peer-to-peer systems tolerant to faults.
Postscript (513K) ]
Last modified on 05/25/2004 by Kris Hildrum.