[Benchmarking] Vector data

[Jeff McKenna]

Comments inline:

Andrea Aime wrote:
> Jeff McKenna ha scritto:
>> Andrea Aime wrote:
>>> GeoServer cannot load a set of separate .shp files as a single entity.
>>> Mind this data set comes from the original RR test thought, I've
>>> never touched it.
>>>
>>> If you think splitting it up in parts will lead to better performance
>>> you should try out that and we can see how much faster things become?
>>
>> Great idea.  In fact I spent some time and generated a new vector test 
>> dataset for the state of Texas using the TIGER 2008 release.  I am 
>> uploading this now so we can all access it:
>>
>> vector-data-tiger08-tx-merged (1.1GB)
>>  - contains merged files for the entire state
>>
>> vector-data-tiger08-tx-counties.zip (1.1GB)
>>  - files stored in individual county folders (as released by the US 
>> Census)

Both TIGER 2008 packages are uploading to the BenchmarkingA server today 
, so we can start using them tomorrow for our scripts.

>>
>> I have updated the benchmarking wiki with descriptions of each 
>> shapefile (http://wiki.osgeo.org/wiki/Benchmarking_2009) and I have 
>> also updated the roads styles page for the new classifications 
>> (http://wiki.osgeo.org/wiki/Texas_roads_styled)
>>
>> My initial tests with MapServer 5.5.0 (dev) are surprising to me:
>> - loading merged shapefiles for entire state (>1 GB) are faster than 
>> by using tileindexes to access individual counties
>>
>> e.g.
>>
>> merged: 
>> http://labs.gatewaygeomatics.com/cgi-bin/wms_benchmarking_by_merged?SERVICE=WMS&VERSION=1.1.1&REQUEST=getmap&layers=vector_benchmarking&STYLES=&SRS=EPSG:4269&BBOX=-96.980629943844,29.854114257812,-96.79038641357,29.974268066406&WIDTH=950&HEIGHT=600&FORMAT=image/png 
>>
>>
>> by county: 
>> http://labs.gatewaygeomatics.com/cgi-bin/wms_benchmarking_by_county?SERVICE=WMS&VERSION=1.1.1&REQUEST=getmap&layers=vector_benchmarking&STYLES=&SRS=EPSG:4269&BBOX=-96.980629943844,29.854114257812,-96.79038641357,29.974268066406&WIDTH=950&HEIGHT=600&FORMAT=image/png 
>>
>>
>> Questions:
>>
>> - what scales should I be focusing on for styling? (or are we only 
>> testing the extents that are listed in the Wiki, of La Grange?)
> 
> We are testing a random set of bboxes generate by Frank's tool,
> but we haven't made any decision on the actual set of scales
> to be tested and the image dimension to be tested.

I could easily spend a week on styling, if we don't agree on certain 
scales to test.  Maybe we can agree on the scales to use at tomorrow's 
meeting?

> 
> So far I've used the attached .jmx and csv files with JMeter
> to make some runs, you may want to have a look on your machine
> and check it out (on the benchmarking server we're more or less
> running blind due to command line usage and the jmeter
> configs being hard to hand edit, so I suggest everybody has
> the ability to run JMeter locally as a GUI app too).
> 
>> - I need some feedback on what other layers I should style (such as 
>> pointlm, arealm)...if no feedback i'll just go ahead and come up with 
>> styles for those
> 
> I think the point layer was being styled by ESRI?
> 
> The poly layer is all yours but I did not find any significant thematic
> styling and I proposed some mails ago ("styling tiger_tracs") to
> use a uniform style for the polygons, and Michal suggested we
> try out a uniform color with partial transparency to check out
> how the two servers do with alpha channel management and png24.
> Sounds good to me.
> 

I'll have the new data ready to use on the BenchmarkingA server for 
tomorrow so we can discuss what layers to use for point/line/poly, and 
who should style what.  Is that acceptable?

-jeff