[OSM-talk] Why place matters, slides from Vanessa Lawrence talk

Artem Pavlenko artem.mapnik at googlemail.com
Fri Jan 11 13:00:07 GMT 2008


On 11 Jan 2008, at 12:23, David Earl wrote:

> On 11/01/2008 11:48, Artem Pavlenko wrote:
>>> Incidentally, this is exacerbated by the lazy rendering rule for  
>>> Mapnik
>> What do you mean by 'lazy' rule?  AFAIK, all available hardware is  
>> working hard day an night :)
>
> I mean the way in which a tile isn't rendered until (after) it is  
> looked at rather than proactively when an area changes.
OK, I think this is right. (better check with Jon)
>
>>> - I was puzzled when someone said to me the other day "why is this
>>> housing estate not connected to the rest of the road network?".  
>>> It was;
>>> but adjacent Mapnik tiles were inconsistent (both laterally and  
>>> by zoom
>>> level - and this wasn't a recently mapped area). You may not  
>>> think this
>>> matters, but I think this is a public face and it causes further
>>> confusion and mistrust.
>> We can certainly improve here. Suggestions ?
>
> I would try three things:
>
> (a) Mapnik works on planet, yes?

No, it works on postgis db which is populated with osm2pgsql from  
planet.
I'm sure you'd agree that using planet directly is not a viable  
solution. I know t at h clients are using APIs but this is hardly a  
solution either. I would go further to suggest that current t at h setup  
uses more bandwidth and cpu (through API usage) then actual uploading  
new data, making API slow and not-user friendly. Have we got some  
stats ?

<joking>
I also worry that amount of heat generated by t at h clients is  
contributing to the global warming
</joking>

> So perhaps use the planet diffs to determine areas which have  
> changed and proactively mark all such areas dirty.

The problem is how to merge planet diff into postgis , I think Jonb  
has done some work/research in this area.
Are there existing tools (osmosis?) that given a planet diff would  
return 'dirty' areas?

>
> (b) for all dirty areas, render at all zoom levels (perhaps down to  
> zoom 12, like osmarender) and do the 8 immediately neighbouring  
> tiles of dirty tiles as well for say zoom 13 or 14 and higher.  
> (Many tiles, neighbouring tiles will be dirty anyway, so this  
> amounts to adding one tile around each group of two-dimensionally  
> contiguous dirty tiles.

Sure, this is certainly possible.
>
> (c) install updated tiles at one go so far as possible.
>

This is how it works already. There is no hidden tiles .
> David

Artem




More information about the talk mailing list