I've talked about this idea with some of you at last SOTM and SOTM-EU. The idea is to have render_list taking advantage of disk cache by rendering tiles in a more "pyramidal" way than the actual full horizontal.
1) better use of disk cache to reduce disk I/O needs Up to now, render_list renders a tile at zoom N but when getting to zoom N+1, most of the needed data are usually the same but not in the disk cache anymore causing a lot of I/O when rendering several zoom level. With my changes, render_list immediately enqueue the 4 underlying N+1 tiles (at the top of the queue). This allows to reduce disk I/O by taking advantage of the caches still containing most of the needed data. That's fine but where should be stop recursion ? 2- pre-render only what's worth prerendering... For this I've added a rendering time limit. If a metatile takes less than this limit, it is not worth pre-rendering it as it will be quickly generated if requested. This allows to limit recursion and at the same time pre-render what can be done fast enough upon request. To use this new feature, I'v only added on option to render_list to control the rendering time limit (in ms). With it, its behavior is unchanged. Example: render_list -m osm -a -f -z 9 -Z 9 --max-time 5000 On my todo list: - when rendering several zoom levels (example: -z 9 -Z 12), check if a metatile has not been already recursively rendered by comparing its timestamp to the render_list start time - add a maximum zoom limit (for the moment recursion can go down to zoom 18) My commit is on github, comments are welcome... before my first mod_tile pull request https://github.com/cquest/mod_tile/commit/652ec90c7d1293a4b049ee0b58be53c01a602e75 -- Christian Quest - OpenStreetMap France
_______________________________________________ dev mailing list [email protected] https://lists.openstreetmap.org/listinfo/dev
