From: Mike B. <gud...@gm...> - 2008-07-10 15:49:48
|
On 7/10/08, Jody Garnett <jga...@re...> wrote: > Yes clipping is probably more expensive then dealing with more content; you > may want to pre-process your data (splitting along tile devisions) prior to > servering it up via geoserver? I have tried this and was able to get almost a 2x performance improvement, however it took 90 minutes to preprocess one layer using java.awt.geom.Area.intersection(). I was also surprised I only got a 2x performance improvement by tiling the dataset. A typical polygon in the dataset has 40,000 vertices. GeoServer seems to handle the large polygons much better than java.awt.geom.Area.intersection(). However, ultimately the performance is no where near the performance of serving up prerendered/pretiled images or polygons. > Depends on the datastore; sometimes we throw out points as we read them (if > say they are off screen; or we already have a point for that pixel). It > really depends on the implementation. I am using Oracle. > I suspect in your case it is returning all the data and not even both to > clip just making use of a graphic context that is the size of a tile. As far as I can tell, that is exactly what it is doing. Evidently java.awt.Graphics2D is much faster at clipping than java.awt.geom.Area. |