[Openstreetmap-dev] ticket 134

[Mikel Maron]

> Were these tiles produced by ruby/api/wms/streets.rbx ?

There's a couple stages. First the request is received by mapserver. For each "layer" (streets, gpx, landsat), it makes another WMS request.

http://www.openstreetmap.org/wiki/index.php/Squid_Usage has a decent outline of the current architecture, and my thought on a potential improvement.

> Where did this program go wrong when the all-white tiles were 
> output?
> 
> Where did this program go wrong when no streets were drawn?

Both requests for streets and landsat timed out. The server is simply overloaded.

> How could we improve this situation?  Or at least find out more 
> about it?  Could the program write "sorry" to STDERR when it fails 
> to do its job, and could I see that logfile, please?

I laid on some thoughts on how to improve it. 

There's one I forgot!! THIS COULD BE CRUCIAL. Apache is also too heavy duty for this job on the tile server. We'd get much better performance replacing it with lighttpd. Anyone have experience with configuring lighttpd?

> Apparently, very similar problems exist in the generation of tiles 
> with yellow marks for the edit applet, which I assume is made by 
> gpx.rbx. The code in gpx.rbx is also very similar to that in 
> streets.rbx.

Yup

> Is the problem in Magick?  Does anything run out of memory or 
> timeout?  What happens?  Which component controls what?

I think mainly Ruby takes too long to do all the calculations and drawing commands.

-Mikel