[Benchmarking] Vector tests completely disk bound now?

Andrea Aime aaime at opengeo.org
Thu Aug 26 07:20:55 EDT 2010

the run I'm making, the first with the new csv files,
seems to hint that the tests are completely
disk bound now: in all the 3 runs against the shapefiles
the disk activity rarely went below 10MB/s (as reported by
"vmstat 1"), normally jumping between 10 and 20MB/s

I also have a PostGIS run in my tests and it shows similar
results: in either of the runs I rarely seen the overall
cpu usage go beyond 5% (as reported by vmstat, where 100%
means all 8 CPU are working full stead).

This is kind of problematic, since if this is true, then we're
king of benchmarking how slow the disk is... (not to mention
the fact a run takes forever to complete).

I'll let it go to see how the raster tests fare, but I'm
wondering if we should change again the csv files to
hit a smaller area: we don't have disks fast enough
for a disk bound test imho (the peak perf at linear
reading is around 60MB/s, but the seek times seem quite

When the test is run it would be good to have someone else
make a similar test and see if they are disk bound too.
MapServer or Mapnik maybe?

Andrea Aime
OpenGeo - http://opengeo.org
Expert service straight from the developers.

More information about the Benchmarking mailing list