Analyzing and Improving the Performance of Spatial Database Processing

Bogdan Simion
University of Toronto
University of Toronto, 2015


   title={Analyzing and Improving the Performance of Spatial Database Processing},

   author={Simion, Bogdan},


   school={University of Toronto}


Download Download (PDF)   View View   Source Source   



Spatial databases have become increasingly important, due to the advent of popular geospatial Web services such as Google Maps, GPS navigation systems, and a host of accompanying location-based services. Spatial databases are used in a variety of real-world applications involving complex data analytics: land surveying, urban planning, environmental assessments, or new BigData application domains like biomedical imaging. As uses of spatial databases become more widespread, there is a growing need for high performance solutions that provide efficient spatial processing. Prior to this work, spatial database processing was not well studied, especially in the current computational landscape. This work focuses on spatial processing in relational database systems, from a performance standpoint. We identify the performance problems involved in spatial data processing, study the nature of spatial queries in terms of resource utilization, and target optimizations for improving performance at both stages of spatial processing. To achieve these goals, we design a spatial benchmark (Jackpine), which provides comprehensive coverage of spatial functionality. We gain insight into spatial workload characteristics by classifying spatial queries in terms of their resource usage, and determine that computations are the main performance bottleneck. Next, we explore avenues for improving spatial query execution, in both the filtering and refinement stages. Given the computational nature of spatial queries, we leverage GPUs for spatial co-processing support in the refinement stage, resulting in considerable speedups. For the filtering stage, where efficient spatial indexes play a crucial role, we take a different approach. We identify the sources of slowdown in a current state-of-the-art RDBMS, which are in great part attributed to unnecessary features and indexing framework abstractions. We propose a new modular data processing framework (Slingshot), which aims to balance performance and generality by drawing abstraction lines differently: decoupling, for each logical database component, the functionality (fixed in the component’s interface) from the implementation (flexible and independently-optimizable for individual application types). We show that overall, Slingshot outperforms traditional RDBMSs, and is comparable to dedicated single-purpose solutions on their target workloads, on both spatial and general database domains. Furthermore, Slingshot’s flexibility allows for easy and efficient integration of GPU support for spatial refinement.
Rating: 1.5/5. From 2 votes.
Please wait...

* * *

* * *

HGPU group © 2010-2024 hgpu.org

All rights belong to the respective authors

Contact us: