[OSM-dev] on-disk indexing of geodata

Freek freek_osm at vanwal.nl
Fri Oct 17 18:47:14 BST 2008


On Friday 17 October 2008, Sascha Silbe wrote:
> ["constant" bulk database with invalidation + "dynamic" database with
> pointer-based indices for updates]
>
> > That might be quite a good idea in OSM practice, you just rebuild the
> > bulk database every so often. Perhaps you can even just use an
> > internal-memory data structure (KD-tree, quad tree, something like
> > that) for the updated part.
>
> Depending on the application, that may also be a useful approach. The
> current mmap() based implementation has the nice side effect that the
> database effectively resides in the OS cache => startup overhead is
> minimal, multiple processes share the data. For a long-running server
> process this might not matter, though.

Looks like we have enough ideas for adding indexes and such to an OSM binary 
format, so if the rest of the structure is fixed I think we can get it off 
the ground (I wouldn't mind at least helping out here and there).

-- 
Freek




More information about the dev mailing list