I got a box with the current hardware config:

CPU: Intel Xeon E3-1220 V2 3.1 GHz
RAM: 10gb
Hard Disk: Western Digital WD10EZRX SATA Hard Drive

OS: Centos 6.5 64 bit
DB: PostgreSQL 9.3.2

and PostgreSQL got following config

shared_buffers = 1024MB
work_mem = 256MB
maintenance_work_mem = 4096MB
fsync = off
synchronous_commit = off
checkpoint_segments = 100

I imported the entire OSM Europe data into PGSQL DB with the following command:

osm2pgsql -c -d gps -U gps --cache 8000 --number-processes 4 --slim europe-latest.osm.bz2

and it took 6 days to complete... it's a lot but it's ok...

now i want to keep my DB updated with the combo of osmosis + osm2pgsql cmds:

osmosis --read-replication-interval workingDirectory=/osmosisworkingdir/ --simplify-change --write-xml-change - | \
osm2pgsql --append -d gps -U gps --cache 8000 --number-processes 4 -s -

Well... this takes MORE than one day to just align single day changes! That makes impossible to keep my local DB aligned!

Am i doing something wrong? Should i upgrade my RAM? Why updating is SOOOOOOOOOO damn heavier than load data for the first time?

3 Answers 3


There's a few issues.

The first one is hardware. Your drive is a WD Green drive, which is generally is about 5400 RPM which is a very slow drive, slower than typical 7200 RPM consumer drives.

One of the biggest tasks in updating is fetching node positions. This is random access, which your drive sucks at.

One option is to use the --flat-nodes option, which will use a 20GB binary file to store node locations instead of a table in the database. This can have significant performance boosts on HDDs as it relies more on sequential access than database tables.

You don't need or benefit from 8000MB memory being used for node cache when updating. 1000 is probably enough. This will, depending on overcommit settings, allow more memory to be used for caching database content.

Your PostgreSQL work_mem and maintenance_work_mem are too high for a machine with 10GB. I'd probably go with work_mem = 64MB and maintenance_work_mem = 1024MB, and even that might be high.

You haven't said where you got europe-latest.osm.bz2, but I'm presuming it's a geofabrik extract. If so, they make daily diffs which allow you to update your database with just data from Europe. To do this, you'd need to edit the osmosis configuration.txt file to change baseUrl to baseUrl=http://download.geofabrik.de/europe-updates. You'd also have to edit state.txt to the right sequence number, which you can find by looking at the dates in http://download.geofabrik.de/europe-updates/000/000/. Once you find the right state file you can download it, e.g. with curl -o /osmosisworkingdir/state.txt http://download.geofabrik.de/europe-updates/000/000/416.state.txt.

Switching to daily diffs from geofabrik should cut back the amount of data that is being added.

Lastly, since you have to reimport for --flat-nodes, you should make sure to download a new Europe extract, and make sure to download PBF, not bzipped XML.

Your new import command should look something like

osm2pgsql -c -d gps -U gps --cache 8000 --number-processes 4 --slim \ --flat-nodes europe_nodes.bin europe-latest.osm.pbf

Your update command would be something like

osmosis --read-replication-interval workingDirectory=/osmosisworkingdir/ \ --simplify-change --write-xml-change - | \ osm2pgsql --append -d gps -U gps --cache 1000 --number-processes 4 --slim \ --flat-nodes europe_nodes.bin -

Of course, the biggest speed boost possible is probably from using an area smaller than all of Europe.

Even with all this, rendering tiles off of a 5400RPM drive is never going to be fast.

  • I'm trying to import pbf instead of the bz2 file + new settings (cache set to 5000 and memory used by postgre conf decreased a lot)... it is from 8 to 15 times faster!!!! i'll post results as soon as it gets completed. meanwhile... thank you Commented Apr 30, 2014 at 15:29
  • Does the 28GB node file (europe_nodes.bin in the example) need to be saved to allow for updates? If not, does keeping it speed up the process? I'd like to just delete it after the import.
    – kontextify
    Commented Aug 18, 2015 at 12:34
  • The flat nodes file is only needed for updates with diffs, so you can safely delete it if not updating. If you're not updating with diffs, you should also add --drop to the command line, which will avoid creating some indexes which are only used during updates. This can significantly speed up imports. If you've already done the import and want to drop the update-only tables, you can drop planet_osm_nodes, planet_osm_ways, and planet_osm_rels. Commented Sep 11, 2015 at 9:12
  • My understanding is that you cannot use --extra-attributes and get the osm_timestamp field when you import from the pbf. Hence, I am using bz2 files exclusively (and the --write-xml-change on the osmosis side).
    – aaryno
    Commented Feb 16, 2016 at 20:41

Without optimizing your toolchain you are lost. Have a look here:


for some details. It`s not only about RAM, SSD instead of SATA hard drives can speed up a lot too.


To add:

  • use latest version of Postgres (now 11). Postgres has performance improvements with each release.

  • please keep fsync=on. synchronuous_commit=off gives the same performance boost, but fsync=off makes your DB non-fault-tolerant.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.