Efficient indexing and view dependent ranking in CFD databases

Methods numerically simulating the interaction of gases or fluids with complex surfaces (computational fluid dynamics, CFD) are able to perform calculations with increasing levels of detail due to the ongoing development of more powerful computers. CFD simulations are utilized during the design of e.g. combustion engines or airplanes, amongst many others. An increasing level of detail on the one hand allows for more accurate and meaningful simulation results proving very useful in industrial development and research. On the other hand, huge amounts of raw CFD data are generated and need to be repeatedly accessed during the subsequent interactive post-processing (e.g. isosurface extraction) by experts in the application domain. The efficiency of post-processing can be significantly increased by the use of virtual reality (VR) technology, letting users immerse into the visualized data sets and extracted features. Interactive post-processing is efficiently performed on data sets stored in main memory, which outperforms secondary storage by magnitudes regarding access times. Large CFD data sets not fitting into main memory thus require efficient secondary storage methods. In this thesis, methods are introduced which appropriately arrange CFD data on secondary storage and allow for an efficient access during post-processing. The efficiency of post-processing is improved by novel view-dependent query methods. The continuous extraction and visualization of partial results in the proximity and direct line of sight of the user allow for a “quick first impression” of the result set. The approaches are enhanced by dynamic aspects, reacting to a user freely roaming the VR environment with immediate alignment of query execution and 3 of the result data stream. For CFD data sets simulated over a span of time, prefetching methods allowing for a dynamic visualization of different time steps are presented. Furthermore, the index supported graphics data server IndeGS is presented, which offers the developed indexing and access methods and can be integrated into arbitrary virtual reality frameworks. IndeGS executes post-processing queries according to a multitude of user parameters and streams the result data to the visualizing component of the VR framework. Relational database management systems (RDBMS) offer comfortable means to integrate user-defined indexes. An improvement of the relational interval tree (RI-tree) is proposed and utilized to enable indexing and efficient view-dependent querying of CFD data in the context of interactive post-processing. This work is concluded with the introduction of novel nearest-neighbor query methods on high-dimensional data. The precalculation of nearest-neighbor information combined with a two-step dimensionality reduction allows for a very high query throughput on static indexes in main memory. This thesis is structured as follows: Part I gives an overview over the topics presented in this thesis. Part II and Part III introduce and evaluate the secondary storage indexing methods and view-dependent query techniques. Part IV presents the approaches to execute view-dependent queries with an enhanced RI-tree in RDBMS. Part V addresses dimensionality reduction methods for efficient nearestneighbor queries. This thesis is finally concluded and aspects for future research are presented in Part VI