[OSM-dev] jxapi performance benchmarking

Paul Norman penorman at mac.com
Wed Oct 12 18:01:19 BST 2011


I recently had the chance to run benchmarking on my jxapi server before and
after a VACUUM ANALYZE; on a database that had been running since April with
AUTOVACUUM off. The DB was imported with osmosis 0.38 and maintained with
osmosis 0.39. With the exception of the initial catch-up and for some time
after server downtimes it has been keeping up with minutely diffs.

 

The server is an AMD Athelon II x3 with a software RAID10 array of 4 7200
RPM consumer SATA drives. The drives are 3x WD Caviar Blue and 1x WD Caviar
Black. This represents about the top end of disk performance before
upgrading to server drives or SSDs. Sequential read speeds are 90-130 MB/s.
iostat reported 40-80% utilization on the physical drives and 300-800 iops
during the VACUUM ANALYZE; The server has 4 GB of ram.

 

Before benchmarking I stopped minutely updates, shut down postgres and
dropped memory caches. The DB was on sequenceNumber=1018433 (Sun Oct 09
15:11:31 PDT 2011)

 

Restarting postgres and dropping memory caches added 1.353s to a 49.75s
query.

 

I ran 10 queries that cover a range of common jxapi queries. These, as well
as the individual times, are listed at
http://wiki.openstreetmap.org/wiki/User:Pnorman/jxapi

 

The VACUUM ANALYZE; reduced the query time by 16.47%. Small queries were not
significantly impacted.

 

The vacuum analyze took approximately 59 hours. I am told this could have
been improved by increasing maintenance_work_mem over the default .25 GB.

 

Faster disks would of course increase all times. More memory might help too.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstreetmap.org/pipermail/dev/attachments/20111012/571ddbdc/attachment.html>


More information about the dev mailing list