Routing website using Gosmore routing engine

An update is running as we speak. It will take about a week before it’s ready (this is what I’m told).

The update was finished on the 28th. I’ve updated the information on the website accordingly.

I do notice that there are some strange routing decisions taken after this update, which I can’t relate to OSM data errors. I’ll relay this to the Gosmore author.

What happens here with YOURS?
from 48.90929 8.64634 to 49.27942 7.23073 using fastest “circles” aorund the target.

Yes, the latest update produced weird results. I’ve moved back to to the update from March 18th. I’ve sent an email to the Gosmore author to inform him of the problem with the update.

Routing data from planet file: 2011-06-28
would you update, please?


I’ve been using for quite some time. But recently I needed a permalink to a route I’ve made, calculated right and as expected, but upon clicking on the permalink I saw a different way around added to the correct one, like this. If I move starting/ending point or tell it to recalculate the route, it shows the correct one, but permalink always adds that strange route around. Sharing such routes will confuse people.
If this problem was reported before, I beg your pardon, didn’t find any references.

Oeps didn’t notice this before. Update is already running for a few days and should finish this weekend.

Looks like a bug where opening the permalink always generates a route for motorcar AND the requested route type. I’ll see if this can be fixed.

Hey Lambertus,

I’m trying to set up a gosmore-based “routing backend” - I guess quite similar to yours - and I’m running into a lot of frustrating little issues, mostly due to missing documentation.

Currently I’m trying to get routing to work for “outside germany”, but creating a .pak file for say europe is impossible on 32 bit hardware (needs more than the addressable 2G Ram), my local system only runs 32bit, and Amazon EC2 is giving me major headaches.

I guess I’ll get it done at some point, and I’ll document how I got it done, but I wonder - how do you build the .pak files, do you split into limited boxes before, and what about “international routing”?

As I need the routing to be able to cross box borders (i.e. from Hamburg to Stockholm), I think there’s a lot of stuff that needs to be done to make this happen?

I’ve read the gosmore rebuild page on the osm wiki, but somehow I can’t make it fit to the current dev state… scratchinghead

Any advise welcome.

I’ve also noted that headless support was broken by some Android changes, but I was able to edit around them and will discuss that with “the author” (I guess Nic) by mail.

I guess a mailing list could or dedicated message board would be really helpful for the project :confused:

The author of Gosmore is indeed Nic Roets. And he also provides the routing backend for since about a year or so. I also had lots of problems trying rebuilds. Some of these problems are due to difference in locale settings. E.g Dutch use ‘,’ as decimal separator while English countries use ‘.’. But this particular bug has been fixed long time ago, there may be other bugs lurking around. A 64-bit machine is very useful while working with OSM data (especially if you use large areas). I’ve switched to 64-bit a long time ago (not only for Gosmore but for the Garmin maps as well), I don’t know if Gosmore is capable of rebuilding a PAK file on 32 bit platforms. Windows won’t work anyway for PAK rebuilding.

For problems with Gosmore I will have to redirect questions to Nic Roets, but you can ask here as Nic reads these forums as well and it will probably help others. Just open a new topic in the Development forum for your project. Btw, there is a low traffic generic osm-routing mailinglist.

So the routing backend is run by Nic himself? I’ll ask him about details on how to build the proper PAK files in that case. Or maybe he stumbles over this post. I also have his mail address (it’s in the sources for Gosmore) so I can contact him directly.

Currently I’m only using the data for private use and for some experiments, so until recently, I did not need more than the german pak file I obtained somewhere in the internet; now, I have to plan a trip to Finland and need more than that :wink:

My EC2 instance is now happily crunching on the europe area, it has finished parsing the XML and is stable at 4.9G RAM, I just go to bed now and hope the best.

I’d use my local MacBook which sure is 64bit, but Gosmore does not compile on that for some Linux dependencies and I just don’t have the time to go fix on that. My older Linux boxes still run on 32 bit, and my web server is shared with other people so I can’t block it completely with such a task.

Yes, since about a year. I was having performance problems on the limited VM it was running. He found a server sponsor and volunteered to manage Gosmore there.

It will probably very difficult technically, as there are so many components that need to work together. But do you think it would be possible to get “social driving” to a routing server? The kind of driving that is found in the app Waze.

I was thinking of the following scheme:

  • a client app (on a smartphone) such as OsmAnd sends statistics to a server about the speed and direction over a short distance.
  • the server processes that speed information to see if there is a problem. In short term, the router will know something (like an accident) happened and wont give routes that pass along that point for a certain time (say 2 hours). In the long term, the routing service will learn from it, knowing that it’s not a good idea to send people to certain streets at 5PM.
  • client apps can use the better routing

Next to the problem of having a lot of instances working together, there is also a problem with the nature of OSM. First of all is, if a way is edited, the already processed data might be worthless (as the way can have a completely different form or length). So the raw (maybe privacy sensitive) data should be kept for a long time, and processed each time somthing changes.

Do you think it would be possible to develop this? Or am I just dreaming?

With a lot of kind help from Nic, I was able to successfully build my own .pak file on Amazon EC2. I’ve created a - hopefully complete - documentation, and wanted to share it here; I hope this does not violate any rules.

I’m currently using EC2 instances for the routing too, but they are quite expensive for that task if you don’t need extreme scaling (where the benefits should start to pay off), so I’m looking to set up a root server for my routing tests as soon as I have some free time.

For building the .pak file, EC2 is quite usable when you have figured out the necessary stuff; I guess it can be done well below $5 for a whole planet file.

This problem was fixed quite some time ago already. When a new route was planned before the results of the previous plan were received, that gave problems in general. The latest version in SVN contains a few other improvements too, but I think did not update for about a year.


There seems to be an issue with the current gosmore.php in the 1.0 api folder, when exporting as GeoJSON.

It is including the traveltime in the properties object, but doesn’t have a comma after the description, which causes the json parsing to fail. I looked at the svn copy of the 1.0 api, and it appears that traveltime doesn’t exist there, so it looks like a minor sync issue between the site and the development code.

Also, for some reason, the destination coordinates appear as the very first entry in the coordinates array when using the dev api version. I haven’t had the time to look at the code to see if I could fix this issue yet, I was just testing to see if it was usable until the 1.0 api was repaired.

Despite these issues, this is a great service, and thanks for all your hard work.


Something is wrong with the filesystem of the Gosmore routing backend server which causes problems with the route calculations. I’ve contacted the administrator and hope it’s quickly fixed.

About 80.000 to 200.000 routerequests are handled each day by this server, so a lot of people are affected unfortunately.

Edit: the filesystem problem has been fixed.

This a call to the owner(s) of two servers with IP address 194.152.x.x. located in Kyrgyzstan leased from Nurtelecom to contact me: osm [at] na1400 [dot] info.

These two servers are each responsible for much more then 10k requests per day but the requests do not provide an application name or contact information in their request headers. A warning header message is now sent back with each route result and I will wait a few weeks to allow the author(s) of the application(s) to contact me. If I don’t hear anything during this time then the IP addresses will be added to the ban-list.

The same problem as before has reappeared:

Edit: problem should be fixed permanently with a kernel update.

Hi Lambertus,

first of all: Thanks for your great work and support here within the forum !

I’ve got a question: Route calculation on seems to be working for most of the routes. Now I tried the following:

  1. Waypoint 1: Known address
  2. Waypoint 2: Known address (but this house is not directly connected to the street)

Any idea why this is not working?

Thanks in advance!

best regards,


Where is that exactly? Sankt-Rochus Strasse is quite common in Germany… It’s possible that the footpaths are newly mapped and not available in the route database yet, if so then Gosmore simply cannot find a street within a certain radius around the marker.