[Benchmarking] Feedback and some activity, Please!

[Marco Hugentobler]

Hi Andrea

Here are csv and jmx files:

https://dl.dropboxusercontent.com/u/45385184/21781.csv
https://dl.dropboxusercontent.com/u/45385184/qgis_av_21781.jmx

 >Seeing what kind of requests you made (zoom level and image size) is 
going to be interesting.

Image 1000 x 600 - 1500 x 900, Zoom level 1:800 - 1:5'000. Zoom level 
has been selected because most of the cadasdral layers are visible on 
that scale.
And yes, image encoding is a very important factor in this benchmark. 
However, I think it is a realistic scenario and image encoding as well 
as palette generation is indeed a very imortant thing for a WMS server.

 >however that really turns the benchmark into a png encoding contest

In UMN as well as QGIS server, the palette generation for png8 is 
implemented as part of the server code (it does not come from a 3rd 
party lib). So I think comparing that in a benchmark makes sense. If a 
test would be driven by rendering speed, you could also say that the 
test measures the performance of the underlying graphic lib.

In any case, I agree that benchmark results are very specific and a 
different test may show a totally different picture.

Regards,
Marco

On 25.06.2013 12:01, Andrea Aime wrote:
> On Tue, Jun 25, 2013 at 11:13 AM, Marco Hugentobler 
> <marco.hugentobler at sourcepole.ch 
> <mailto:marco.hugentobler at sourcepole.ch>> wrote:
>
>     >So is Stefan shares the data and qgis project, can you share the
>     load test itself? I imagine it's some sort of jmeter thing?
>
>     Yes, it is basically a copy of the foss4g benchmark jmeter file,
>     only with layers / styles replaced (and outputformat is png8).
>     Sampling extents were generated with python script from foss4g
>     benchmark.
>
>
> Can you share it? Or, at least, can you share the command line used to 
> generate it?
> Seeing what kind of requests you made (zoom level and image size) is 
> going to be interesting.
>
> I understand the usage of png8 to some extent, however that really 
> turns the benchmark into a png encoding contest,
> unless rendering is very slow I'd estimate 50-70% of the time is spent 
> building the palette and encoding the output
> PNG (I took the FOSS4G 2010 benchmarks recently and made some 
> optimization to GeoServer, now 75% of the
> time is spent encoding the PNGs (plain ones, no png8) if I use the 
> fastest PNG encoder available, and it gets
> way worse if I leave it with the standard Java PNG encoder. Now... 
> imagine adding on the fly palette computation
> on top of that :-)
>
> 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
>
> -------------------------------------------------------
>
>
> _______________________________________________
> Benchmarking mailing list
> Benchmarking at lists.osgeo.org
> http://lists.osgeo.org/mailman/listinfo/benchmarking


-- 
Dr. Marco Hugentobler
Sourcepole -  Linux & Open Source Solutions
Weberstrasse 5, CH-8004 Zürich, Switzerland
marco.hugentobler at sourcepole.ch http://www.sourcepole.ch
Technical Advisor QGIS Project Steering Committee

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/benchmarking/attachments/20130625/3b03ea87/attachment.html>