Hi, the server at nominatim.openstreetmap.org currently serves 400 request/s and is expected to max out at 500 req/s. So in principle it is possible with a single server.
If you are using the latest release, then it still runs with non-persistent connections to the database. At this line: https://github.com/openstreetmap/Nominatim/blob/627a487fcfd325f8a340f10298a961d15d2760b6/lib/db.php#L6 change $bPersistent to true per default. That should improve performance quite a bit. With latest master you can also reduce the database size for your use case. The smaller the size the more of the database you can keep in RAM. For infos on that feature see http://nominatim.org/release-docs/develop/admin/Import-and-Update/#choosing-the-data-to-import That said, tuning the servers to get the throughput does require quite a bit of administration knowledge. Also consider simply buying a couple of machines and distributing the load. Nominatim can be run with WAL replication or by setting up independent servers. kind regards Sarah On Mon, Mar 25, 2019 at 07:57:06PM +0300, Mark Davies wrote: > Hi All > > I'm currently in the process of trying to improve the amount of geocodes > per sec > > I have the whole world on nominatim but only currently using a handful of > countries > Its running on 48gb ram and SSD's and is topping out around 62 per second > with max CPU > I know it needs at least 96gb ram however my question is regarding scaling > and if its even possible to get near what I want > > My target is 500 per second. > I'm first wondering if this is a possibility, even with 128+ram > I know there are many factors like similar cached requests etc, but someone > might know if there is a limit on what nominatim can handle so I can adjust > my expectations. > > Also if there is a better solution for reverse geocoding with road speed > data (extra_tags) > Thanks! > _______________________________________________ > Geocoding mailing list > [email protected] > https://lists.openstreetmap.org/listinfo/geocoding _______________________________________________ Geocoding mailing list [email protected] https://lists.openstreetmap.org/listinfo/geocoding
