Martijn van Oosterhout wrote:
> On Jan 5, 2008 7:06 PM, Sebastian Spaeth <[EMAIL PROTECTED]> wrote:
>>> - When the server has the list of blank tiles to insert, it could run
>>> one pass over the list to see if it's mostly sea of mostly land and
>>> whatever it is, insert a blank tile with that type at zoom-12 and then
>>> only insert the other types.
>> Yes it could do that. My idea was to not take that hit at each insert,
>> but rather run a cleanup script periodically that does these kinds of
>> optimizations with lower priority. Once a z12 tile is marked as blank
>> land/sea by such a script subsequent blankness entries at higher zoom
>> levels become noop's.
> 
> Sure, but it also cuts the number of insert to be done in the first
> place, which produces a measurable speed improvement.

Which is good, I concur.

  > I actually have a script which grabs all the data for a given set of
> zoom levels and actually calculates the minimum number of blank-tile
> entries needed to represent that. It produces rather spectacular
> results (>90% removal of entries). It's just complex enough that I've
> left it for a few days so I when I look at it again I can recheck the
> assumptions, but I'm pretty sure it works.

Would be great if you could check that in if you are sure that it works. 
There is a trade off between the number of entries and the number of 
look-ups required. Do we wan to remove all unneccessary entries or do we 
  want to keep all, say, z12 entries to make z17 lookups not too slow?

> The way it's setup is a sort of daemon that monitors for newly
> uploaded tilesets and optimises them straight away. It's quite quick
> so eventually I'll probably run it over the entire DB and see how it
> goes. If you're interested I can check it into SVN, maybe you can
> check it for bugs...

would be good. let me know if and where you check it in.

spaetz

_______________________________________________
Tilesathome mailing list
[email protected]
http://lists.openstreetmap.org/cgi-bin/mailman/listinfo/tilesathome

Reply via email to