[Tilesathome] tiles at home = mission impossible?

[Christopher Schmidt]

On Mon, Apr 14, 2008 at 01:46:39PM +0200, Frederik Ramm wrote:
> Hi,
> 
> > That was a rerender of all (at that time existing!) tiles and before the
> > design of lowzoom changed. If you would try this now, you would have to
> > wait for more than 2 years. The design specs for lowzoom now say: We need
> > everything (independent of the existence of data).
> 
> But even if the lowzoom process triggers re-rendering, it doesn't 
> actually refuse to create a z8 tile just because one of the z12 tiles is 
> missing, or does it? So why do you say "have to wait"?
> 
> And as for the t at h design: The bottleneck is downloading data from the 
> API. It would be a waste to download the data and then only render a 
> little bit, and come back later to render the rest (requiring a new 
> download).

I'm not sure that the API is really the blocker anymore. Recently, there
has definitely been a server side backlog of tile processing, causing
our clients to sit idle because the server can't keep up. 

Additionally, in dense areas, there is a definite lack of client
capacity: you can see it in the 'lull' last night, because I uploaded a
bunch more data around Mass.: the end result is that complex tilesets
were being taken by all clients, and the server side (both API and
tah-wise) was mostly idle. 

Also, we now have osmxapi, which is highly performant, staying within
minutes of the API: tests last night indicate a data turnaround from
upload -> availability of less than 8 minutes. So if the 'main' API is
relaly the bottleneck, there is relatively little that can't then be
solved by switching to osmxapi for many use cases.

(I understand all these are somewhat tangential to your point.)

Regards,
-- 
Christopher Schmidt
MetaCarta