Hi Daniel, thanks for your advice - I've done exactly the same way you as you described it for some years now (doing the painful "average"-resampling due to preprocessing the overviews with GDAL and nesrest in WMS-postprocessing). The servers and services are "highly optimized" and generally fast - thanks to GDAL's and mapserver's functionality and performance and community hints. But I wasn't really satisfied with final image-quality, because of the trade-of between performance and image-quality on WMS-postprocessing - I decided for performance.
From my point of view it still makes a difference in image quality, whether I use nearest or average on WMS-postprocessing, especially the aliasing-effects on edges of buildings, bridges, etc. would be nearly totally eliminated with "average"postprocessing , I did some tests! My guess was, that resampling could be done much better by the GPU (having a look in the world of gaming, where some "super-mega-ultra-antialiasing" is done with more than 100 frames per seconds in highest resolutions. Now there is an option to get some new servers with integrated GPUs (Nvidia GTX1080) and asked myself (and now the community ;-) ) whether it would make sense to exercise that option... Regards Ben -----Ursprüngliche Nachricht----- Von: mapserver-users [mailto:[email protected]] Im Auftrag von Daniel Morissette Gesendet: Donnerstag, 1. Juni 2017 16:05 An: [email protected] Betreff: Re: [mapserver-users] OpenGL on Linux (Debian/Ubunu) There have been some experiments a few years ago but I don't remember how far the implementation went, how much real testing it got and whether it is still maintained. You may want to look at http://mapserver.org/development/rfc/ms-rfc-50.html and search for discussions on the dev list around 2008-2011 for the full history. But for your specific problem, have you considered simply generating overviews in your source images using average resampling? This way you would pay the price of (average or bilinear) resampling only once and your WMS requests could use nearest neighbor on the overviews to provide much better quality results without the processing cost. More info about optimizing raster access is available here (in case you had not seen it yet): http://mapserver.org/input/raster.html#preprocessing-rasters Daniel On 2017-06-01 9:35 AM, Schepers, Benjamin wrote: > Hi list, > > has anyone of you made some experience with mapserver and GPU-rendering on > Linux (Debian preferred)? > > The use case would be to speed up raster-resampling on the fly (nearest <-> > average) in WMS and maybe later vector-rendering on complex symbology. > For now raster-rendering in average-mode is three-times slower than > resampling with nearest neighbor (which looks worse, indeed). > > Are there some pitfalls? Maybe compiling with openGL-driver, other issues, > etc.... > > Would GDAL benefit from using GPU on processing very large rasters > (aerialimage-BigTIFFs). How would this be done? OpenGL / OpenCL? > Any hints are welcome, I couldn't find that much info on this... > > Regards > Ben > > _______________________________________________ > mapserver-users mailing list > [email protected] > https://lists.osgeo.org/mailman/listinfo/mapserver-users > -- Daniel Morissette Mapgears Inc T: +1 418-696-5056 #201 _______________________________________________ mapserver-users mailing list [email protected] https://lists.osgeo.org/mailman/listinfo/mapserver-users _______________________________________________ mapserver-users mailing list [email protected] https://lists.osgeo.org/mailman/listinfo/mapserver-users
