[OSM-dev] on-disk indexing of geodata

Freek freek_osm at vanwal.nl
Fri Oct 17 14:25:59 BST 2008


On Friday 17 October 2008, Sascha Silbe wrote:
> Actually it seems I've been limited in view by my own application. The
> problem I'd have with updating data isn't how to map data updates to
> tree updates (the topic covered by the paper mentioned above) but how to
> efficiently do inserts and deletes on an on-disk tree. 

Generally you keep a bit of free space in all nodes to accommodate a number of 
insertions and deletions, and only split or merge nodes when they overflow or 
underflow so that you don't need to move around all kinds of elements all the 
time.

> I haven't done 
> any research on that topic, though, since I don't need it yet (bulk
> imports are quite fast - ~25 minutes for the 865MB europe.osm.bz2).

That's not bad, I suppose your index still fits in memory? I think that in 
most cases the top part of the tree should fit in memory and only the leaves 
(containing pointers/addresses to the actual items) will stay on disk during 
operation.

-- 
Freek




More information about the dev mailing list