Efficient retrieval and scalable storage of multi-dimensional data.

摘要

Databases increasingly integrate different types of information such as multimedia data. As a result, it is becoming necessary to support efficient storage and retrieval of multi-dimensional data. In several modern database applications, both the dimensionality and the amount of data that needs to be processed are increasing rapidly. Therefore, it is important to develop techniques that overcome the scalability and the dimensionality problems of multi-dimensional data sets. Since the amount of data is large, it is crucial to develop techniques that exploit parallelism in large-scale databases. In this context, we propose partitioning and declustering techniques for multi-disk architectures. Several effective solutions for the high dimensionality problem are also proposed: access structures for efficient searching, and dimensionality reduction techniques to remove the curse of dimensionality. In particular, we propose a compression based index structure, a clustering based approximate search technique, and a dimensionality reduction technique using inner product approximations. Finally, we discuss two new types of queries and propose efficient techniques to process them. Extensive experimental evaluation of all presented techniques has been performed and comparison with other state-of-the-art approaches is presented.