yoric / mozilla-paris-talks Goto Github PK

With the growth of internet traffic, various systems have proposed
peer-to-peer or peer-assisted content distribution. However, most existing
models use flooding or structured distributed hash tables to provide a
given node a view to the whole content available in the network.
While arguable, this design leads to long lookup times, and
thus prevents the use of these models for applications where speed
matters more than the amount of available content. Here, we introduce
behave, a new peer-to-peer overlay, build on top of gossple's gossip
based similarity clustering. Behave records the interactions of users with the
network, like web-site visits or domain-name requests, and builds
implicit profiles that reflect user interests. Nodes compare their profiles
with those of other peers in order to select communication partners that
may offers the most interesting content according. Using this small subset
of the network, the node can perform efficient lookups in one hope for the content
the user has high probability to be interested in. Such a network,
by allowing faster lookups for interesting parts of the content, is
by design especialy suitable for building distributed caching
systems like web or DNS caches.