Brief announcement: Trilix: a scalable unstructured lookup system for dynamic environments

Unstructured lookup systems such as Gnutella incur small constant overhead upon each single join or leave operation and can easily support keyword searches. Search in such systems is not structural, and may fail. However, queries usually succeed in locating files since popular files are held by many nodes. Thus, such systems are suitable for deployment over the Internet. One of the main challenges in unstructured lookup systems is to construct and maintain a good overlay topology. First of all, it is desirable to construct a balanced overlay, where all nodes have roughly the same role and degree, and the load on all nodes is equal. This is becoming more important now, as many ISPs have started to limit the maximal bandwidth consumption of every user. Furthermore, it is desirable that each node will have a constant degree, independent of the number of nodes in the system, in order to achieve high scalability. Secondly, in dynamic failureprone environments, it is desirable to have a robust overlay that remains connected despite failures. A third important property of peer-to-peer lookup overlays is captured by the notion of duplication ratio: when a query is flooded over a subset of the overlay, the duplication ratio is the probability that a search message reaches a node that has previously received it [3].