Prepare raster data for wms using geoserver!

Question

I have some orthophoto of different resolution (urban 8cm pixel, hill 20cm pixel, mountain 35 cm pixel) which cover an area of 28000 km2. Those are in jpg format, already georeferenced.

I need to build a WMS with this amount of data. I need some best practice based on geoserver for building this service. Server requirements, raster format to use etc...

One other thing while working with this is how can i avoid raster overlaps of different resolution for certain area. Because for the same are could be more than one raster with low resolution or higher resolution. I need to get the higher resolution which is for that particular area.

Answer 1

Step one is to read GeoSolutions excellent GeoServer on Steroids: http://demo.geo-solutions.it/share/foss4g2013/gs_steroids_sgiannec_foss4g2013_01.03.pdf - it's pretty much the definitive guide to GeoServer optimisation. You'll want pages 10-18.

A simplified version of what you want to do is:

• Convert into GeoTiffs
• Use Inner Tiling
• Add Overviews
• Load them as an ImageMosaic.

It's a fairly straightforward task with GDAL, as documented in that PDF.

If you do that then you're unlikely to need any special Server Requirements etc unless you're anticipating very high load, in which case you may want to turn it into a WMTS/TMS (tiled) product rather than use WMS.

This GeoServer list thread may also be of use: http://osgeo-org.1560.x6.nabble.com/Tips-on-serving-Orthophotos-as-ImageMosaic-tt5093083.html#a5093104%3E

Answer 2

I have just begun using TileMill in conjunction with GeoServer and it's pretty great.

In TileMill, you can add all of your rasters and style them/define at which zoom levels they will be visible. Then export the project to a single .mbtiles file. You can define what range of zoom levels to build tiles for, also.

You'll need to install the MBTiles extension for GeoServer. Once you do that, you can use the single .mbtiles files as a data store in GeoServer. Doing this will give you a single layer in GeoServer which may or may not be desirable. However, that layer can contain all of the original rasters, because, as defined in TileMill, only certain ones will show up at certain zoom levels.

I don't know how this method holds up with a very high volume of data, but in theory it seems sound.