Distributed Intersection Join of Complex Interval Sequences

In many different application areas, e.g. space observation systems or engineering systems of world-wide operating companies, there is a need for an efficient distributed intersection join in order to extract new and global knowledge. A solution for carrying out a global intersection join is to transmit all distributed information from the clients to a central server leading to high transfer cost. In this paper, we present a new distributed intersection join for interval sequences of high-cardinality which tries to minimize these transmission cost. Our approach is based on a suitable probability model for interval intersections which is used on the server as well as on the various clients. On the client sites, we group intervals together based on this probability model. These locally created approximations are sent to the server. The server ranks all intersecting approximations according to our probability model. As not all approximations have to be refined in order to decide whether two objects intersect, we fetch the exact information of the most promising approximations first. This strategy helps to cut down the transmission cost considerably which is proven by our experimental evaluation based on synthetic and real-world test data sets.