GPU Accelerated Web Map Services

08/02/2013 - 4:35am to 4:55am
ML - Main Seminar Room
Matthew Grossman


Web Map Services (WMS) are a well-established technology used for managing and delivering the basemaps and supporting imagery used in combination with geographical data.  There are three data and compute-intensive tasks involved in serving this imagery: decoding the source data, resampling the source data to the requested scale and extent, and compositing resampled data from many different source files into a seamless image.  Parallelism can be exploited on all three fronts, making these computationally intensive steps particularly well suited to implementation on the GPU.  In this project we develop a WMS implementation that uses a GPU-based solution to these tasks, with the ultimate goal being the development of a fully parallel pipeline of tasks that are handed off to the GPU for processing.  We then develop comparisons between several methods of implementation, including a purely CPU-based implementation.