[OSM-dev] Creating Easy Map Embedding

Andy Allan gravitystorm at gmail.com
Mon Oct 22 16:27:41 BST 2007


On 10/22/07, Christopher Schmidt <crschmidt at metacarta.com> wrote:
> On Mon, Oct 22, 2007 at 03:01:46PM +0100, David Earl wrote:
> > Why don't we create a distributed tile server, so the load can be spread
> > around? Then we do have "infinite resources like Google".
>
> Building a distributed tile server is a lot of work, and not easy. (I
> say this having built the innards of a distributed tile cache for
> TileCache.)
>
> How many people have the ability to set up and maintain an up to date
> copy of the mapnik database? How many people have the server resources to
> deploy mapnik, PostGIS, etc. as well as maintaining an up to date copy
> of the planet that doesn't fall behind? I think that if you browse
> informationfreeway, which uses distributed caches of tiles, you'll see
> the benefit this has, but also the drawback, which is that you're
> depending on your tile caches to always be available -- when they
> aren't, the whole map suffers.
>
> If the answer to 'how many people can bring the same level of resources
> to bear on the problem as tile.openstreetmap.org' is 'few', then
> distributed tile servers don't get you much, because all the rendering
> still has to be done in one place, and (once mod_tile is deployed)
> rendering is a lot more work than actually serving tiles.
>
> In other words: once the mod_tile solution goes out, and a 10x increase
> in performance is available, distributed tiles don't actually improve
> the situation much/any.
>
> A distributed tile cache is a difficult problem to solve, and it's not
> the most important part of the problem here: again, mod_tile will
> probably solve the short term performance needs of serving mapnik tiles
> from the OpenStreetMap servers. (I could be wrong on this, but this is
> the impression I've gotten.) If there is a desire to work on a
> highly robust distributed tile serving system, that's great, but it
> should be after mod_tile is completed and replaces the current mapnik
> tile serving setup.
>
> The more important question -- at least to me -- is "What resources does
> OSM want to be providing to users of OSM?" Is 'tiles' one of them, or is
> the OSM tile server only a resource designed to be used for OSM.org, and
> everything else belongs off site? I don't know how others feel about
> this question, and I'd be interested in hearing what others think.

I think distributing the tiles between people / organisations is a bad
idea. I think the informationfreeway front page is the best example
why - bits keep dropping off whenever I'm looking at it.

What problem are we trying to solve? Bandwidth is effectively
unlimited at UK HE sites (to a certain extent - we have a gig backbone
uplink here at work which is being upgraded to 10G at some point but I
suspect that's an order of magnitude more than we'll need any time
soon), but that's only from a technical point of view. I think our
hosts at UCL may get annoyed at some point well before that e.g.
constant 100Mbps. If that's the limiting factor, *then* we'll need to
look at hosting somewhere else, and hope someone comes with stacks of
gold to pay for it.

If it's the load on the machines, then we would be much better off
with tile1 - tile5 all sitting behind a load balancer and under the
control of the same person in the same facility.

But I think a reasonable setup would be to offer free use of
tiles.osm.org bandwidth up to a certain monthly limit, at which point
we would expect third parties to host (if not render) their own tiles.
So small websites have a low barrier of entry, and those big enough
for us to notice are asked to provide their own resources.

Cheers,
Andy




More information about the dev mailing list