Hi, > For simple map calls there is TRAPI[1]. As far as I know, TRAPI performs > much better on map (bbox) queries than either the main-API, XAPI or ROMA (on > equivalent hardware). Rather than using a database, I think it used a > pre-tiled file structure, so that it simple needs to peace together a bunch > of tiles, rather than do a full search in a db, which massively reduced the > disk seeks necessary. You point out several points which are correct but not very well known. TRAPI exists, feeds the tiles@home clients, uses a pre-tiled structure directly in the file system and is rather fast. Not clue why it is ignored so often.
> I don't think XAPI is all that much better suited for map requests than the > main API and imho it was rather unfortunate that the mantra has been for > people getting banned from the main API for scraping to use XAPI. This just > bogged down XAPI for people who actually needed its extra capabilities. Right and that is why the tiles@home clients talk to the TRAPI load balancer instead of the XAPI. But to be honest most people scraping the XAPI would be better served by using a planet dump or extract thereof instead of arduous scraping a web service, be it XAPI or TRAPI. Actually I have been thinking of creating a patch to jXAPI that answers every large map call from an IP that has recently requested several other large map calls with an empty OSM file containing a comment along the lines of "hey you. don't bog down this server but grab an extract". Patrick "Petschge" Kilian. _______________________________________________ talk mailing list talk@openstreetmap.org http://lists.openstreetmap.org/listinfo/talk