Am Freitag, den 30.03.2018, 12:38 +0200 schrieb Philippe Mouawad:
> ok for me.
> Make your dream become reality :)

I have started on inserting the HdrHistogram into StatCalculator, but
that is not really possible, as StatCalculator is written as a generic
and HdrHistogram (as I have understood it) best with long values.

Currently we have two non-generic implementations of StatCalculator.
One with Long and one with Integer. The Integer one seems to be used by
a test, only.

Is the Integer version used somewhere else, that I have overlooked?

At the moment I am trying to insert HdrHistogram into
StatCalculatorLong and delegate all calls to it. But I believe it would
be better to insert HdrHistogram into StatVisualizer and
SamplingStatVisualizer, since StatCalculator is public interface and I
don't see a way to keep it backward compatible with my current
approach.

What do you think?

Felix

> 
> On Friday, March 30, 2018, Felix Schumacher <
> [email protected]> wrote:
> 
> > 
> > Sorry for the late answer.
> > 
> > As https://bz.apache.org/bugzilla/show_bug.cgi?id=61725 is yet
> > another
> > report for the same "bug" and HdrHistogram seems to be a good
> > compromise on memory, speed and accuracy, I think we should go that
> > way.
> > 
> > HdrHistogram is available under BSD clause 2, so it should be safe
> > to
> > include.
> > 
> > As a side effect I am dreaming of a mini-histogram in the
> > visualizer
> > for every request line :)
> > 
> > Felix
> > 
> > Am Dienstag, den 21.11.2017, 14:33 +0530 schrieb Isuru Perera:
> > > 
> > > +1 for using HdrHistogram. We found out that the percentiles
> > > values
> > > shown
> > > in "Aggregate Report" and the web dashboard are different for the
> > > same JTL
> > > file. It will be good if we can use the same logic to calculate
> > > percentile
> > > values in JMeter listeners and the web dashboard.
> > > 
> > > On Sat, May 20, 2017 at 7:58 PM, Philippe Mouawad <
> > > [email protected]> wrote:
> > > 
> > > > 
> > > > 
> > > > Hello,
> > > > Any other thoughts ?
> > > > 
> > > > Thanks
> > > > 
> > > > On Saturday, May 6, 2017, Antonio Gomes Rodrigues <ra0077@gmail
> > > > .com
> > > > > 
> > > > > 
> > > > wrote:
> > > > 
> > > > > 
> > > > > 
> > > > > +1
> > > > > 
> > > > > 2017-05-06 15:57 GMT+02:00 Maxime Chassagneux <
> > > > > [email protected] <javascript:;>>
> > > > > :
> > > > > 
> > > > > > 
> > > > > > 
> > > > > > +1 : lgtm
> > > > > > 
> > > > > > 2017-05-06 15:41 GMT+02:00 Philippe Mouawad <
> > > > > [email protected] <javascript:;>
> > > > > > 
> > > > > > 
> > > > > > > 
> > > > > > > 
> > > > > > > :
> > > > > > > 
> > > > > > > Hello,
> > > > > > > A user recently reported a bug on StatCalculator class:
> > > > > > > 
> > > > > > >    - https://bz.apache.org/bugzilla/show_bug.cgi?id=61071
> > > > > > > 
> > > > > > > It appears issue is not restrained to Median but
> > > > > > > concerns  percentile
> > > > > > > computing.
> > > > > > > 
> > > > > > > Besides, this class has many drawbacks:
> > > > > > > 
> > > > > > >    - It is not tested as much as it should be
> > > > > > >    - It relies on an algorithm that consumes memory
> > > > > > > 
> > > > > > > We use another class in Web/Dashboard report and
> > > > > > > BackendListener
> > > > client
> > > > > 
> > > > > 
> > > > > > 
> > > > > > 
> > > > > > > 
> > > > > > > 
> > > > > > > implementations that relies on commons-math
> > > > > > > org.apache.commons.math3.stat.descriptive.DescriptiveStat
> > > > > > > isti
> > > > > > > cs.
> > > > > > > 
> > > > > > > My proposal is the following:
> > > > > > > 
> > > > > > >    - First step : Introduce in StatCalculator,
> > > > > > > HdrHistogram
> > > > > > >    - Second step : Replace DescriptiveStatistics by
> > > > > > > HdrHistogram
> > > > > > > 
> > > > > > > 
> > > > > > > Benefits:
> > > > > > > 
> > > > > > >    - Uniform computing accross JMeter
> > > > > > >    - Better performances
> > > > > > > 
> > > > > > > Are you ok with this approach ?
> > > > > > > 
> > > > > > > Thanks
> > > > > > > --
> > > > > > > Regards.
> > > > > > > Philippe Mouawad.
> > > > > > > Ubik-Ingénierie
> > > > > > > 
> > > > --
> > > > Cordialement.
> > > > Philippe Mouawad.
> > > > 
> > > 
> 

Reply via email to