Distributed Search with Rendezvous Search Systems

The ever-growing amount of data made available through the Internet requires powerful and highly scalable search systems. Not only the mostly unstructured data but also the increasing user expectations demand for sophisticated and flexible search primitives like keyword search or XPath. Academic research on peer-to-peer networks has been mostly focused on how to build such advanced primitives on top of the keyvalue-lookup primitive offered by Distributed Hash Tables (DHTs) [2], [10]. While in most cases possible, this approach often requires expertise in both distributed systems and information retrieval beyond the typical application developer. Rendezvous search systems take a completely different approach by separating the matching of queries and data from the network communication. This not only allows to use the same rendezvous search system unchanged for many different search primitives, but also to easily integrate existing non-distributed implementations of search primitives. The application developer provides the rendezvous search system a data storage backend and a match function to determine matching data from the backend for an incoming query and the rendezvous search system takes care of the distribution of data and queries in the distributed environment. In my talk I want to give a short overview of the current state of the art of rendezvous search systems. The name rendezvous search system refers to the guarantee that a query meets a copy of every data item in the system eventually. How to fulfill this guarantee is up to the rendezvous search system and depends on the target environment, but typically requires O( √ n) copies of queries and data respectively.