[OSM-talk] OpenStreetMap Pledge

[Tom Carden]

On 30/05/06, Christopher Schmidt <crschmidt at crschmidt.net> wrote:
> On Tue, May 30, 2006 at 10:36:51PM +0100, Tom Carden wrote:
> >
> > It wasn't a rhetorical question.  I meant why can't *your* implementation
> > run on OSM's servers?
>
> Because the implementation I've used, and found to be successful, has
> been rejected by the OpenStreetMap project -- specifically, using
> geographic file formats and rendering them using MapServer. I was told
> this was because "Mapserver is too slow" for OSM's needs.
>

That sounds familiar.  I believe Steve found it too slow when he
experimented with it, but I believe he also expressed concerns about
its extensibility with regards to running openstreetmap.org on it in
the long term.

> What I believe is more likely is that mapserver was being used in a way
> that it was not as efficient as possible, and therefore, was acting more
> slowly than one would expect.

And also because the current database is set up to be read/write and
record all changes.

Taking planet.osm, putting it in a database, and rendering it as tiles
on the fly is what Nick did with PHP/GD.  Now he's using RubyMagick.
As far as I know, there's no cache behind his demo code, and the only
bugs are to do with the dev server setup.

The reason openstreetmap.org is slower is because it takes all the
changes into account and shows the current data (which Nick's code
doesn't do, and which your Mapserver implementation couldn't and
wouldn't need to do).  I believe it will be possible in the future to
optimise the database so that queries for "most recent" (which will
grow faster than writes) become a lot faster, but I've not studied the
database set-up in detail yet.

> In the past month I've been working with
> OSM data, I've watched tile rendering times for my OSM based maps
> decrease fivefold to twentyfold, as I learned more about how to make
> things fast.

Sounds great! Please share!

>
> It seems to me that OSM does not want to go the route of using
> shapefiles and mapserver to render tiles. I understand that. However,
> that's exactly the route I want to go, and I'm perfectly willing to
> share benefits -- if there turn out to be any -- with the OpenStreetMap
> project. But until I can set up something that does the same -- or
> better -- than what OSM does now, I can't convince anyone that this is
> actually useful.
>

True that.

> In other words, once I *have* an implementation, I'd be glad to run it
> on OSM's servers -- if it was desired. Until I do, there's no point in
> me wasting anyone's time with a solution that's already been rejected.

Is this like, "I have discovered a way to render OpenStreetMap data
quickly and correctly using Mapserver, but the margin is too small to
write it here"?

> Since the solution in question has already been rejected, it's likely
> that members of the OSM project will consider it a waste of time -- so
> I'm offering to compensate them for that waste of time monetarily, if it
> will help motivate action.
>

Fair enough.  I'd be surprised if it does, but I won't rule it out.

Tom.