[Wps-discuss] WPS Benchmarking at FOSS4G 2015

Mon Jul 6 16:21:17 PDT 2015

Thinking about this more, it changes one problem (implementations are
dependent on their geospatial library for speed) with another
(implementations are dependent on their xml parsing / format encoding
technology for speed).

So not sure of the value here, other than a discussion point :)

--
Jody Garnett

On 30 June 2015 at 15:50, Jody Garnett <jody.garnett at gmail.com> wrote:

> I am happy with any approach that gives us a chance to both talk about our
> implementation(s) and promote WPS. We could also use the opportunity to
> discuss what we are looking forward to in WPS 2.0 etc...
>
> GeoServer may be a bit hampered with the echo process idea (as its
> strength is on processing data served up by the services) but that is fine
> ... always room for improvement.
>
> --
> Jody Garnett
>
> On 23 June 2015 at 06:40, Benjamin Proß <b.pross at 52north.org> wrote:
>
>>  Dear WPS developers,
>>
>> As I mentioned in a previous mail, we originally planned a
>> interoperability testing for this years WPS benchmark.
>> I had a discussion with my colleagues lately and we came to the
>> conclusion that we will not have results ready for the FOSS4G in September.
>> As a plan b, I think we could re-use the test concept from last year, as
>> it was quite sophisticated.
>> However, we had some discussion during the preparation for the last
>> benchmark about the execute tests.
>> I brought up that we could use a simple echo process, which would have
>> some benefits:
>>
>>    -  we could easily test execute with all three data types
>>    -  we are independent from the libraries used by the different WPS
>>    implementations for geospatial computations (i.e. we are testing only the
>>    performance of the WPS)
>>    -  it could be re-used for WPS 1.0/2.0 compliance tests at the OGC
>>
>> Some points regarding the echo process from earlier mails:
>>
>> Am 19.08.2014 um 16:03 schrieb Benjamin Proß:
>>
>> We do not yet have any results to compare but we think that benchmarking
>> with "real" processes that compute something could falsify the results, as
>> the computation in most cases is done by a underlying library (GeoTools in
>> our case, but this could also be exchanged).
>> So the results would be influenced by how fast the geo-computation lib
>> would do its job. Not sure that we want that?!
>> We also do not test whether the process really did its job (e.g. buffer
>> the input geometry). So do we need to use "real" processes?
>> We would propose to use an echo process that could have three inputs,
>> Literal-, Complex- and BBoxData. The process then simply returns what it
>> gets (input/output types would have to match).
>> We would probably also need a asynchronous version that has a delay or
>> something.
>> That way we could test the input-/output handling capabilities (of all
>> three data types) of the services and we would still test everything
>> regarding execute that is tested right now (i.e. we would not loose
>> functionality).
>>
>>
>> Am 21.08.2014 um 11:16 schrieb Fenoy Gerald:
>>
>> I think it is a great idea to handle testings of the Execute request the way you described, by using a simple echo service.
>>
>> By now for synchronous requests, the tests look like the following:
>>
>> 	• Test 1: value as JSON
>> 	• Test 2: value as GML
>> 	• Test 3: value as reference as GML
>> 	• Test 4: same as test 3, result as reference
>> 	• Test 5: value as reference as GML using POST request
>> 	• Test 6: same as test 5, result as reference
>>
>> And for asynchronous request, we have the following:
>>
>> 	• Test 1': value as reference as GML
>> 	• Test 2': same as test 1, result as reference
>> 	• Test 3': value as reference as GML using POST request
>> 	• Test 4': same as test 3, result as reference
>>
>> Personally, I think that the Test 1 and Test 2 should become unique now as there is no need for parsing any value from the service itself as it will simply return the value without any treatment. So we should use only the Test 2 for our next run.
>>
>> I also suppose that we should multiply the number of tests as we will need to ask for each available outputs individually and all the possible combinaisons of them to also test this cases.
>>
>> So, I would like to propose the following updates for the tests run for both synchronous and asynchronous requests. We run each individual test 7 times: first to have only once output at a time (so 3 tests), then to test all possible combinaisons of 2 outputs names (3 tests) and finally to have all the output returned (1 test).
>>
>> I hope that you can confirm that I’m right thinking of this updates to the current tests.
>>
>> I’ve added one column to the table [1] listing all WPS team participating to write the name of the echo service when it will be available.
>>
>> Best regards,
>>
>> [1] http://wiki.osgeo.org/wiki/WPS_benchmark_2014#Available_infrastructure
>>
>>
>> What do you think about this approach?
>>
>> Cheers,
>>
>> Benjamin
>>
>> --
>> Benjamin Proß
>> Software Engineer
>> 52°North Geoprocessing Community
>>
>> 52°North Initiative for Geospatial Open Source Software GmbH
>> Martin-Luther-King-Weg 24
>> Fon: +49-(0)-251–396371-42
>> Fax: +49-(0)-251–396371-11b.pross at 52north.orghttp://52north.org/
>>
>> General Managers: Dr. Albert Remke, Dr. Andreas Wytzisk
>> Local Court Muenster HRB 10849
>>
>>
>> _______________________________________________
>> Wps-discuss mailing list
>> Wps-discuss at lists.osgeo.org
>> http://lists.osgeo.org/mailman/listinfo/wps-discuss
>>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/wps-discuss/attachments/20150706/afe6e97d/attachment.html>