[OSM-dev] Postgres 8.4 performance trouble in osm2pgsql setup

Emilie Laffray emilie.laffray at gmail.com
Fri May 7 10:08:59 BST 2010

Could be also Linux related. Phoronix has been talking about regression due
to some modification in ext4. It could be due to this too.

Emilie Laffray

On 6 May 2010 23:14, "Frederik Ramm" <frederik at remote.org> wrote:


   I had posted about performance problems with Postgres 8.4 in
September last year already. With Ubuntu Lucid just released, which
comes with 8.4 by default, I played around some more. Unfortunately I am
still a little bit away from having systematic, comparable results, but
I have the impression that importing updates with osm2pgsql is somehow
buggy (performance-wise) in 8.4.

I'm running this on a machine with SSD. With 8.3, the process was
disk-bound despite SSD, which is what I would expect, after all there's
not much to compute.

With 8.4, a (large, ~ 1 day) test diff takes about four times as long to
load. This is unacceptable and I will probably have go back to 8.3
because of that until I can find out exactly what the problem is.

When searching for possible reasons on the web, I found this message
about Postgres 9.0 and irregularities with intarray index usage:


It could have *something* to do with what I am seeing, however I could
not reproduce intarray operations being slower from within a function
than from the command line.

I'll investigate this further and report back with anything I find; for
the time being just be warned that there are isolated indications of
Postgres 8.4 being less suitable for OSM than 8.3. (I am largely
ignorant of any changes between 8.3 and 8.4, and simply set my config
options to the same I had been using with 8.3; it is possible that 8.4
requires something to be configured differently to work.)

I'd also be interested to hear from anyone who, despite this message,
plays with 8.4 (or even 9.0 alpha) and has something to say.


Frederik Ramm  ##  eMail frederik at remote.org  ##  N49°00'09" E008°23'33"

dev mailing list
dev at openstreetmap.org
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstreetmap.org/pipermail/dev/attachments/20100507/9fe24236/attachment.html>

More information about the dev mailing list