[OSM-dev] Proposal: Database accelerator and dirty tile marker based on simple algorithm.

Nick Hill nick at nickhill.co.uk
Sat Sep 16 22:42:15 BST 2006



Etienne wrote:
> On 9/16/06, *Nick Hill* <nick at nickhill.co.uk 
> <mailto:nick at nickhill.co.uk>> wrote:
> 
>     Nick Hill wrote:

> 
> Currently most detail is recorded to about 10m accuracy which is good 
> enough for roads.  It is possible (a long time in the future) that  we 
> will be recording detail at 1m, 10cm or even 1cm accuracy.  This will be 
> good enough to record the outline of residential houses, street 
> furniture and the like. 
> 
> If you use zoom level 0 as the lowest level of detail then you would 
> need to introduce zoom levels -1, -2 etc to cater for increased detail.  
> If you call the very highest level 0 and work down then you wouldn't 
> have this problem.

This is consistent to what we have now, so would presumably be best.

> 
> Likewise the alpha-quadtree proposal, because it is variable length, 
> seems to me to be infinitely extensible in the detail direction.  I'm 
> not sure that the 32-bit approach would be so easily extensible.

We are talking about two things to which the tile relates.

1) The invalidation tile. There is a balance between invalidation tile 
size and invalidation database size for smaller invalidation tiles. a 
655m invalidation tile in data terms is very convenient, and appears to 
be a practical size.

2) Selecting objects based on which tile they are in and what zoom level 
they should be shown at. The table which is queried would only have 16 
bits for each lat and lon. The object itself would hold further 
information of it's exact location. The tile would hold 16 bits lat + 16 
bits lon. The object would hold 16 bits lat + 16 bits lon, providing a 
total granularity of 1cm or better.






More information about the dev mailing list