I am trying to optimize globaly the performance of our OsgEarth based application.
During tests, I observed a big difference on memory consumption depending on the rendering technic.
Here is the test:
- OsgEarth version is 2.8 tag
- Fill a earth file to only display one big shape file (world boundaries from natural earth at 10m precision) using feature_geom driver
-> memory consumption in case of "drape" is 255 MB
-> memory consumption in case of "map" is 161 MB
I understand that the drape technic is using a texture (4096 in my case) and a RTT pass.
However, I don't really understand why the memory takes 100 MB more when using drape technic (the "mesh" sould be present only once in the memory)
Any reasons? is it an issue?
If not, is there a work around for decreasing memory concumption in this kind of situation?
A 4096 RTT texture takes up 64MB of memory; I did a quick test with the default 2048 size and saw my memory footprint go up about 24MB (expected 16MB from the texture). So, there is some extra memory usage happening; not sure yet why that is.