Hello,
The 0.1 seconds is specific for the query on the ALL_OBJECTS system table...
The queries doing real work take typically indeed a fraction of this time...
average about 0.03 seconds in our case... but those are layers containing
between 5 and 5 million objects... Calling the same query returning all 500
objects of a table is definitely not representable in any case as this will
just be cached... and starting from oracle 11R2 it will most likely even be
cached in a client-side resultset cache in the JDBC driver...
We just tried upgrading the jdbc driver to 11.2.0.3 (same version as the
database)... and this doesn't solve the problem :(...
Pieter
Van: [email protected] [mailto:[email protected]] Namens Andrea Aime
Verzonden: vrijdag 25 oktober 2013 12:30
Aan: Pieter Roggemans
CC: [email protected]
Onderwerp: Re: [Geoserver-users] Oracle vector JNDI: bad performance due to 16
database calls per layer fetched
On Fri, Oct 25, 2013 at 12:18 PM, Andrea Aime
<[email protected]<mailto:[email protected]>> wrote:
On Fri, Oct 25, 2013 at 10:27 AM, Pieter Roggemans
<[email protected]<mailto:[email protected]>>
wrote:
In the overview above you see that 2), 5) and 6) is the same query being
executed 3 times per layer... so 30 times in our average case. If there is a
"SCHEMA" defined (any) in the JNDI data store, this query takes 0.1 second per
execution in our config... if not... +- 1 second per execution. This means that
best-case this query accounts for 0.1 * 30 calls = 3 seconds added time for
each WMS request... which means 66% of the time because our average WMS request
takes 4.5 seconds.
Btw, the 0.1s per query caught my eye. I have a small Oracle table in a vanilla
Oracle XE installed on Ubuntu 12.04 64 bit,
that I'm displaying fully (500 multilines), and I did a quick performance check
making 1000 wms requests with "ab" (ApacheBench), using a single thread.
Here is the result:
Requests per second: 14.24 [#/sec] (mean)
Time per request: 70.203 [ms] (mean)
While I was at it, I've turned down logging to PRODUCTION setting and installed
my new fast PNG encoder (available as a community module
for the dev series, but works fine in 2.3.x as well), and gave it another kick:
Requests per second: 18.27 [#/sec] (mean)
Time per request: 54.736 [ms] (mean)
The queries are definitely taking less than 10ms each on my machine (as
confirmed by the v$sql AVG_ELAPSED_TIME_S column)
Cheers
Andrea
--
==
Our support, Your Success! Visit http://opensdi.geo-solutions.it for more
information.
==
Ing. Andrea Aime
@geowolf
Technical Lead
GeoSolutions S.A.S.
Via Poggio alle Viti 1187
55054 Massarosa (LU)
Italy
phone: +39 0584 962313
fax: +39 0584 1660272
mob: +39 339 8844549
http://www.geo-solutions.it
http://twitter.com/geosolutions_it
-------------------------------------------------------
------------------------------------------------------------------------------
October Webinars: Code for Performance
Free Intel webinars can help you accelerate application performance.
Explore tips for MPI, OpenMP, advanced profiling, and more. Get the most from
the latest Intel processors and coprocessors. See abstracts and register >
http://pubads.g.doubleclick.net/gampad/clk?id=60135991&iu=/4140/ostg.clktrk
_______________________________________________
Geoserver-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/geoserver-users
