Moving Observer Support for Databases

Interactive visual data explorations impose rigid requirements on database and visualization systems. Systems that visualize huge amounts of data tend to request large amounts of memory resources and heavily use the CPU to process and visualize data. Current systems employ a loosely coupled architecture to exchange data between database and visualization. Thus, the interaction of the visualizer and the database is kept to the minimum, which most often leads to superfluous data being passed from database to visualizer. This Ph.D. thesis presents a novel tight coupling of database and visualizer. The thesis discusses the VR-tree, an extension of the R-tree that enables observer relative data extraction. To support incremental observer position relative data extraction the thesis proposes the Volatile Access Structure (VAST). VAST is a main memory structure that caches nodes of the VR-tree. VAST together with the VR-tree enables the fast extraction of appearing and disappearing objects from the observer's view as he navigates through the data space. Usage of VAST structure significantly reduces the number of objects to be extracted from the VR-tree and VAST enables a fast interaction of database and visualization systems. The thesis describes other techniques that extend the functionality of an observer aware database to support the extraction of the N most visible objects. This functionality is particularly useful if the number of newly visible objects is still too large. The thesis investigates how to optimize a given observer path. We propose query load balancing strategies that depend on a step-size and an incremental slice size. The proposed strategies balance a number of queries issued to the database and ensures that none of objects is overstepped along the path. All solutions have been implemented and evaluated empirically. A number of experiments validate the performance, scalability, and effectiveness of the optimizations.