Hacker News Reader

Comment by lucianof | Hacker News Reader

When I first looked at OpenStreetMap a couple years ago I dismissed it solely because I didn't like the tile design. But now I see more and more really beautiful designs popping up (e.g. also those from http://mapbox.com/) and I expect OSM to gain a lot of importance in the next years (overtaking Google Maps maybe?). These 3rd-party tile designs a beautiful example of how much more can be done with data that's truly open!

The one thing I don't understand is who pays for hosting it. Once everyone uses OSM, won't it be prohibitively expensive for a non-profit to serve it? Maybe I'm wrong and it's not that much of an issue. Does anyone have an idea how much it costs Google to serve the Google Maps API? Wikipedia is in the same situation, and they solve it with their donation pleas. Maybe it would make sense if a fraction of our ISP bill automatically goes to the most visited non-profit websites. They offer kind of a public good to everyone, after all.

View on HN

Disclosure: I work for MapBox.

Two parts: OpenStreetMap, the website, aims mainly to be a database and a great editing interface. So the end result is mostly the data, which you can download in full ( http://planet.openstreetmap.org/ ). They have a few sponsors to handle the bandwidth required ( http://wiki.openstreetmap.org/wiki/Partners ), and a few meaty servers ( http://munin.openstreetmap.org/ ). There could always be more, and the base of partners could easily increase.

Part two is that _tiles_ and design are a different matter - OSM.org itself has low limits on the number of tiles you can use from their server before they start pushing you to use your own server or another service. MapBox is one of those services, and basically we handle a massive amount of work & bandwidth, and price out the service based on how much these things cost.

Once everyone uses OSM, won't it be prohibitively expensive for a non-profit to serve it?

Yes. The problem is that Google has started charging now as well- at least, once you hit a certain threshold. They might have a deal available for non-profits, I'm not sure. It's certainly something I'd hope MapBox are working on, because non-profit mapping seems like a large (and worthwhile area).

No doubt, the bandwidth costs are expensive, though. I made my own custom map tiles (you can see a demo of them in use at http://www.taxono.my) and put them behind an Amazon CloudFront server, which makes it about as cheap as possible. But still, if my traffic gets huge, it could be an issue.

> (...) which makes it about as cheap as possible.

Have you tried Cloudflare?

"Everyone" won't ever use OSM - its license is prohibitively restrictive for a lot of uses.

It's great for anyone that don't want to combine it with data they can't release, but for anyone that needs to mash it up with proprietary data, it's a non-starter by design. I think a lot of people don't realize this, as I've seen more than one company use OSM data in ways that either puts them in conflict with the license or will force them to open data they've previously been very protective of...

But as others have noted, most people will need to host themselves or pay someone to host it for them once they reach certain thresholds.

(disclosure: I've got an interest in a company in the GIS space that can't use OSM data exactly because of the license, so I do have a vested interest in it)

Could you point to the part of the OSM license that says this?

I think you can layer data on top of tiles from osm with the openlayers api without licensing issues.

You're right that you can do probably that if you only ever keep the layers separate. The moment you e.g. provide export functions that combine the layers, you're currently in a grey area. With the new license, it's no longer a grey area, but you explicitly need to release your source data to be in compliance.

Not as long as the other items have no dependency on or interaction with the OSM layer.

There is nothing that restricts the display of a proprietary data layer on top of an OSM render : in the resulting mashup, the variously licensed sources of data each remain in its own layer and there is therefore no mixing of conflicting licenses.

On the other hand, if you produce new data by processing OSM data or mixing it with your own, then you are constrained by the bounds of Openstreetmap's license.

> On the other hand, if you produce new data by processing OSM data or mixing it with your own, then you are constrained by the bounds of Openstreetmap's license.

.. such as the moment you provide a function to generate PDF versions of the maps, or any number of other features that end up mixing the data. So in effect you either accept severely limiting your features, or you use other sources of map data.

No - as long as the other items have no dependency on or interaction with the OSM layer, the generated PDF is clear of license conflict: it is still merely a composite display of independent sources, not a work produced on the basis of OSM data. If the OSM data was rendered without modification then it is not necessary to release the other data sources. Cf. http://wiki.openstreetmap.org/wiki/Open_Data_License/Use_Cas...

They've changed their tune then, because that used to be explicitly called out as an example of a grey area in the old license that they would be prevented with the new license.

If that's correct then that's fantastic.

Given that it's contradictory to what used to be claimed, it still makes me concerned about touching it without having a lawyer going over the license in great detail, though...

Once everyone uses OSM, won't it be prohibitively expensive for a non-profit to serve it?

Wikipedia says no, but that may be an outlier. IMO they should start internalizing costs (especially for embedded maps) now before they get out of control, perhaps by throttling access so that high-volume users are encouraged to mirror and serve the data themselves. Coral CDN has a similar strategy: you can use it for free, but only a certain amount.

It's not clear if the 'they' in your statement is referring to OSM or a hypothetical non-profit, but, if OSM, they've always done this, I believe.

There are plenty of big users of OSM based tiles that would easily take down the OSM servers, so they have to either host their own or use some third-party to do so (MapQuest is surprisingly popular here, since they have capacity to spare and they'll serve OSM tiles up to some limit before starting to charge).

"OpenStreetMap data is free for everyone to use. Our tile servers are not."

http://wiki.openstreetmap.org/wiki/Tile_usage_policy

Ah, I see they're ahead of me. But that policy is pretty vague about what heavy use means.

If you are a commercial provider, you can expected to be held to tight standards regarding your definitions. A not-for profit gets a lot more leeway, and "heavy use" is about as specific as OSM needs to get. If you are wondering whether you are crossing the line from "reasonable" to "heavy" use, just assume you've crossed it and make an effort to move your traffic onto your own tile server.

Does anyone know how much disk space a map tile set of the whole world takes up? It obviously depends wholly on zoom levels, etc, but if anyone has done this themselves I'd love to hear more about their experience.

about 54 TB. But only about 2% of tiles have ever been seen. Each zoom level has 4 times as many tiles as the previous level and zoom level 18 (the highest) doesnt have a lot of data.

Very interesting numbers - would you happen to have a source for them ? Is that the traffic for Openstreetmap.org's Mapnik tiles ?

http://wiki.openstreetmap.org/wiki/Tile_Disk_Usage

Quote: "OSM does NOT pre-render every tile. Pre-rendering all tiles would use around 54000GB of storage. As the following table shows, the majority of tiles are never viewed. In fact just 1.79% are viewed."

Those figures are about a year old, though it does give a newer figure for space used which is about 1200TB or 2.7% of the total size estimated.

Mapbox's Streets global basemap is ~240 GBs.

source: http://macwright.org/presentations/nodedc/#7

OpenStreetMap pays to host tiles that are intended to be used by mappers. Ideally you would use one of the other providers (MapQuest Open, MapBox, make your own, etc.) if you are releasing a product that doesn't directly relate to improving the OSM map data.

Reply on news.ycombinator.com