[Qgis-developer] Approximate reprojection for vectors

[Marco Hugentobler]

Hi Martin

>In any case we should look into the performance in a more coordinated way:
>1. choose a dataset for testing - ideally just 1-3 layers of real data
>with lots of features (e.g. 10000+) for each feature type - points,
>polylines, polygons. Any ideas for a free dataset?
>2. write a benchmark that could be run repeatedly easily - I would
>volunteer to provide it
>3. test and analyze the results
>4. optimize bottlenecks

I like the idea of having a performance test suite. Ideally there should be 
the possibility to add new datasets / tests (map rectangles, symbologies) to 
it.

> Is the benchmark (data + configuration) publicly available? Is it
> simple to download it and run on a local machine?

The information for the last two benchmarks is here:
http://wiki.osgeo.org/wiki/Benchmarking_2011
http://wiki.osgeo.org/wiki/Benchmarking_2010

The infrastructure is there during the year, so we can always make test runs. 
The configuration for 2011 is not ideal in my opinion, too complicated with 
over 100 layers and some test arrangements are a bit strange (e.g. lots of 
very small rectangles in one test).

Regards,
Marco

Am Dienstag, 4. Oktober 2011, 17.53:19 schrieb Martin Dobias:
> On Mon, Oct 3, 2011 at 3:22 AM, Marco Hugentobler
> 
> <marco.hugentobler at sourcepole.ch> wrote:
> >> the time spent within postgres provider while waiting for server to
> >> return the features may not be considered.
> > 
> > That might be a valid point. Is it better to use oprofile for the waiting
> > time?
> 
> OProfile is a system-wide statistical profiler. Being system-wide
> means that it profiles all applications (and kernel, too!) running on
> the system. So it is able to measure time not only in qgis but also in
> postgresql etc. And being statistical means that it is periodically
> triggered by hardware and records samples what code is running at each
> time. With enough samples you should have quite good idea what is
> going on.
> 
> However I am not yet sure how to profile i/o waiting times - maybe
> strace is a better tool in this regard.
> 
> Additionally there is no such nice GUI like kcachegrind for oprofile
> output - although op2callgraph tool may be of some use.
> 
> > In any case, the other servers in the benchmark had to fetch the features
> > from the same database. So there must be very performance critical
> > differences in the other parts of the code. As you might know, QGIS
> > didn't really win first price...
> 
> Is the benchmark (data + configuration) publicly available? Is it
> simple to download it and run on a local machine?
> 
> Regards
> Martin