[OSM-dev] osm2pgsql slim mode, postgis, and hard disk spindles

Stephan Knauss osm at stephans-server.de
Fri Sep 4 23:59:08 BST 2009


Hi Frederik,

Frederik Ramm wrote:
> scenario, and then applied the 20090819-20090820 daily diff with 
> osm2pgsql --slim in a number of different configurations:
> "plain" with -C4000.......................... 262 minutes


Can you post some details of your setup, please? My machine takes about 
double the time for a 24h update.

It's a slightly different configuration but that shouldn't change things 
a lot.

I use osmosis to pipe in the cumulated changes of the hourly diffs. I 
modified the rulset to discard "created_by" and added another name-tag.

Maybe there are more points to optimize the postgres setup than 
mentioned in the wiki.

My machine is quite new, I expected a bit better performance.
As already mentioned an average run is about 7 hours.
CPU is an i7 920 with 12G RAM. Postgres running on Win x64.
By looking at the performance monitor I assume the system being IO-bound.

The disk system was not tailored to be hosting a huge database like 
this, but I expected better performance being a RAID-setup. Sequential 
read is around 400MB/s.

I could also provide some iometer results, but don't know what 
parameters to use to get representative results. How many threads to 
configure? What percentage of random IO?

Some real-world measurements from other running machines would be great.

Stephan




More information about the dev mailing list