[Geoserver-users] WMS performance/data source question

[Not sure if my original post went through, sorry for the repost if it did]

We've successfully set up a rather large and complex set of shapefiles
(about 8GB total, some files >1GB not memory mapped) that are served
by Geoserver as a WMS.

However, we are wondering, are shapefiles the most efficient data
source, performance wise? Or should we look into using PostGIS, or
something else?

Regards,
David

David E. Reksten ha scritto:

[Not sure if my original post went through, sorry for the repost if it did]

We've successfully set up a rather large and complex set of shapefiles
(about 8GB total, some files >1GB not memory mapped) that are served
by Geoserver as a WMS.

However, we are wondering, are shapefiles the most efficient data
source, performance wise? Or should we look into using PostGIS, or
something else?

It depends :slight_smile:
If you are using GeoServer 1.6.0-rc2, and you're not using significant
attribute filtering, shapefiles should be some 2-10 times faster than
Postgis (and Mysql as well) when you're displaying the whole data
set, and more or less on par if you're displaying a small part of it,
provided geoserver managed to create the spatial indexes.
For shapefiles this big geoserver may not be able to (see
http://jira.codehaus.org/browse/GEOT-1498) but you may be able
to create them using the MapServer shptree utility.

If, on the other side, you're doing significant attribute filtering
(stuff like, display only major roads at a certain scale, that is,
you're using attributes to reduce the amount of geometries to be
displayed), then postgis is to be preferred.
Of course a dbms is the preferred solution also if you're going to
use WFS-T against your data.

Ah, if you're using GeoServer 1.5.x series or older, I warmly
suggest you upgrade to 1.6.0, shapefile rendering is a lot faster there.

Hope this helps
Cheers
Andrea

Andrea Aime ha scritto:

David E. Reksten ha scritto:

[Not sure if my original post went through, sorry for the repost if it did]

We've successfully set up a rather large and complex set of shapefiles
(about 8GB total, some files >1GB not memory mapped) that are served
by Geoserver as a WMS.

However, we are wondering, are shapefiles the most efficient data
source, performance wise? Or should we look into using PostGIS, or
something else?

I forgot to add one thing. Certain options for the jvm running GeoServer
make a lot of difference performance wise. See the "GeoServer in production" guide:
http://docs.codehaus.org/display/GEOSDOC/6+GeoServer+in+Production+Environment

Cheers
Andrea

On 10/01/2008, Andrea Aime <aaime@anonymised.com> wrote:

Andrea Aime ha scritto:
> David E. Reksten ha scritto:
>> [Not sure if my original post went through, sorry for the repost if it did]
>>
>> We've successfully set up a rather large and complex set of shapefiles
>> (about 8GB total, some files >1GB not memory mapped) that are served
>> by Geoserver as a WMS.
>>
>> However, we are wondering, are shapefiles the most efficient data
>> source, performance wise? Or should we look into using PostGIS, or
>> something else?

I forgot to add one thing. Certain options for the jvm running GeoServer
make a lot of difference performance wise. See the "GeoServer in
production" guide:
http://docs.codehaus.org/display/GEOSDOC/6+GeoServer+in+Production+Environment

Thanks. Our current setup is Geoserver 1.6.0-beta2 running standalone
webserver with jre1.6. We're only serving static data, so WFS-T is not
necessary. We've used shptree were necessary, as well. I think we'll
have to start playing with SLD transparency and antialiasing next. Our
Geoserver jre setup is already setup to use about 500MB of stack
space, which seems a bit much, but at least it's working quite well
that way.

I had a look at
http://www.foss4g2007.org/presentations/view.php?abstract_id=120 which
was very helpful and answered a lot of questions (although I really
think it should be made available as a PDF, I had to install
OpenOffice just to read it -- email me to get a pdf I made out of it).

Looking forward to see how much we're able to squeeze out :slight_smile:

Regards,
David