Abstract:

A major problem within peer-to-peer systems is to find the best
peer given a certain query. Inspired by the work in the area of
social networks we present a novel peer-to-peer system called
INGA (Interest-based Node Grouping Architecture). Peers
cooperate to efficiently route queries along adaptive shortcuts
based overlays using only local knowledge. We propose active and
passive shortcut creation strategies and a new routing algorithm
that combines a greedy, high degree and flooding based search
depending on one’s knowledge. We quantify the benefit of the
overlay network by comparing the performance of INGA in the SWAP
simulation infrastructure against a simple Gnutella style network,
against recently proposed shortcut networks without similarity
metrics and against a relaxation based approach. While obtaining
the same recall we show in our experiments that with INGA we half
the messages for a query.

Date: 2004-Dec-09     Time: 16:00:00     Room: 336

For more information:

• http://algos.inesc-id.pt/algos/reading.html