performance

[Michael Heinrichs]

Hi Johan,

I would be careful with using a benchmark app. The problem is that there is no typical JavaFX application and any kind of performance optimization must be heavily weighted against possible performance degradation for other types of applications. A benchmark app tends to skew your optimization effort towards optimizing mostly for the benchmark app. Instead I would extract benchmarks from applications where you see performance issues, e.g. I would try to extract a benchmark from the application where you noticed the slower swipe events, and optimize only until the performance becomes acceptable. This will still distort your optimization effort, but at least towards a real world issue. :)

But I am pretty sure you are aware of the risks and want to use a benchmark app anyway. ;) A good suite of benchmarks is GUIMark 2 http://www.craftymind.com/guimark2/ <http://www.craftymind.com/guimark2/>. I think there is a JavaFX implementation, but I am not sure, if it is available. What is missing completely in GUIMark 2 are UI controls, which are the most important ingredient in many applications. I remember one very simple, but highly effective benchmark: just a ListView with many entries, which was scrolled up and down. (That particular use case was always killing us on mobile.) Now I would probably use a TableView or TreeTableView.

When I investigated poor response time, I usually measured the time between triggering an event until the screen changed as a reaction. In my opinion, it does not make sense to measure an input event in isolation, because poor response time can be caused by something completely different, thus one should try to get as close as possible to what the user experiences.

Cheers,
Michael



> On 20 Feb 2015, at 16:47, Johan Vos <johan at lodgon.com> wrote:
> 
> Hi,
> 
> I am looking for a "standard" way to test "performance" of JavaFX on
> different platforms (especially mobile). I know there are a number of apps
> that show the fps, and it can also be measured in prism itself, but is
> there some benchmark app?
> 
> Also, is there a quantitative standard for measuring the quality of how a
> JavaFX implementation responds to input events? The reason I'm asking the
> latter is because on iOS, the response on swipe events is "slower" than on
> Android -- but that's not a quantitative statement.
> I am confident we can improve this, but it would be nice to have some
> before and after numbers, to measure the improvements.
> 
> Thanks,
> 
> - Johan