You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@jmeter.apache.org by Philippe Mouawad <p....@ubik-ingenierie.com> on 2019/07/11 19:07:49 UTC
Percentiles: Using LatencyUtils or IntCountsHistogram
Hello,
Context:
- Working on switching percentiles computation from custom+DescriptiveStats
(commons-math) to HDRHistogram
- Vladimir suggested using LatencyUtils as it provides detection of Pauses
- It appears LatencyUtils's class to use is LatencyStats, which uses
AtomicHistogram
which uses Long to count, which means this has a bigger footprint than for
example IntCountsHistogram (which uses Long for values , but int for counts
which is enough in our context)
Question 1 :
- Does it look to you ok to stay on LatencyUtils ? for now we built our
work on it
Question 2:
- Javadocs of LatencyStats#recordValue mentions nanosec, but it looks like
we could use directly millis, can somebody confirm ?
If some of you want to look at work in progress of PR:
- https://github.com/ubikloadpack/jmeter/pull/50
References:
- https://groups.google.com/forum/#!topic/mechanical-sympathy/C6d-vFeYs_c
-
https://hdrhistogram.github.io/HdrHistogram/JavaDoc/org/HdrHistogram/AtomicHistogram.html
-
https://github.com/LatencyUtils/LatencyUtils/blob/d8f51f39f6146e1ad9a263dc916bcbc0ec06e16d/src/main/java/org/LatencyUtils/LatencyStats.java#L196
Thanks
Regards
Qian Qian and Philippe
--
[image: logo Ubik Ingenierie] <https://www.ubik-ingenierie.com> Philippe
Mouawad
Senior Performance Expert
320914981 <+33320914981> | p.mouawad@ubik-ingenierie.com
[image: ubik-ingenierie.com] ubik-ingenierie.com
<https://www.ubik-ingenierie.com> | [image: 03.20.91.49.81] 03.20.91.49.81
<+33320914981> | [image: 23 rue du chemin de fer , 59100 , Roubaix] 23 rue
du chemin de fer, 59100, Roubaix
<https://www.openstreetmap.org/#map=18/50.69454/3.16455>
Re: Percentiles: Using LatencyUtils or IntCountsHistogram
Posted by Philippe Mouawad <p....@ubik-ingenierie.com>.
Hello,
Thinking more about using LatencyUtils, I wonder if it's a good idea to use
it in the end as it seems there has been no activity on the project since
december 2015.
So, if we decide not to use it, we'll switch to IntCountsHistogram.
I was hoping to have some help here, as I prefer to have at least 2 brains
on this subject.
Thanks
Regards
On Tue, Jul 16, 2019 at 8:49 AM Philippe Mouawad <
p.mouawad@ubik-ingenierie.com> wrote:
> Hello,
> Any feedback ?
>
> Thanks
>
> On Thursday, July 11, 2019, Philippe Mouawad <
> p.mouawad@ubik-ingenierie.com> wrote:
>
>> Hello,
>>
>> Context:
>> - Working on switching percentiles computation from
>> custom+DescriptiveStats (commons-math) to HDRHistogram
>> - Vladimir suggested using LatencyUtils as it provides detection of
>> Pauses
>> - It appears LatencyUtils's class to use is LatencyStats, which uses AtomicHistogram
>> which uses Long to count, which means this has a bigger footprint than for
>> example IntCountsHistogram (which uses Long for values , but int for counts
>> which is enough in our context)
>>
>> Question 1 :
>> - Does it look to you ok to stay on LatencyUtils ? for now we built our
>> work on it
>>
>> Question 2:
>> - Javadocs of LatencyStats#recordValue mentions nanosec, but it looks
>> like we could use directly millis, can somebody confirm ?
>>
>>
>> If some of you want to look at work in progress of PR:
>>
>> - https://github.com/ubikloadpack/jmeter/pull/50
>>
>>
>> References:
>> - https://groups.google.com/forum/#!topic/mechanical-sympathy/C6d-vFeYs_c
>> -
>> https://hdrhistogram.github.io/HdrHistogram/JavaDoc/org/HdrHistogram/AtomicHistogram.html
>> -
>> https://github.com/LatencyUtils/LatencyUtils/blob/d8f51f39f6146e1ad9a263dc916bcbc0ec06e16d/src/main/java/org/LatencyUtils/LatencyStats.java#L196
>>
>> Thanks
>> Regards
>> Qian Qian and Philippe
>> --
>>
>>
>> [image: logo Ubik Ingenierie] <https://www.ubik-ingenierie.com> Philippe
>> Mouawad
>> Senior Performance Expert
>> 320914981 <+33320914981> | p.mouawad@ubik-ingenierie.com
>> [image: ubik-ingenierie.com] ubik-ingenierie.com
>> <https://www.ubik-ingenierie.com> | [image: 03.20.91.49.81]
>> 03.20.91.49.81 <+33320914981> | [image: 23 rue du chemin de fer , 59100
>> , Roubaix] 23 rue du chemin de fer, 59100, Roubaix
>> <https://www.openstreetmap.org/#map=18/50.69454/3.16455>
>>
>
>
> --
>
>
> [image: logo Ubik Ingenierie] <https://www.ubik-ingenierie.com> Philippe
> Mouawad
> Senior Performance Expert
> 320914981 <+33320914981> | p.mouawad@ubik-ingenierie.com
> [image: ubik-ingenierie.com] ubik-ingenierie.com
> <https://www.ubik-ingenierie.com> | [image: 03.20.91.49.81] 03.20.91.49.81
> <+33320914981> | [image: 23 rue du chemin de fer , 59100 , Roubaix] 23
> rue du chemin de fer, 59100, Roubaix
> <https://www.openstreetmap.org/#map=18/50.69454/3.16455>
>
>
--
[image: logo Ubik Ingenierie] <https://www.ubik-ingenierie.com> Philippe
Mouawad
Senior Performance Expert
320914981 <+33320914981> | p.mouawad@ubik-ingenierie.com
[image: ubik-ingenierie.com] ubik-ingenierie.com
<https://www.ubik-ingenierie.com> | [image: 03.20.91.49.81] 03.20.91.49.81
<+33320914981> | [image: 23 rue du chemin de fer , 59100 , Roubaix] 23 rue
du chemin de fer, 59100, Roubaix
<https://www.openstreetmap.org/#map=18/50.69454/3.16455>
Re: Percentiles: Using LatencyUtils or IntCountsHistogram
Posted by Philippe Mouawad <p....@ubik-ingenierie.com>.
Hello,
Any feedback ?
Thanks
On Thursday, July 11, 2019, Philippe Mouawad <p....@ubik-ingenierie.com>
wrote:
> Hello,
>
> Context:
> - Working on switching percentiles computation from
> custom+DescriptiveStats (commons-math) to HDRHistogram
> - Vladimir suggested using LatencyUtils as it provides detection of Pauses
> - It appears LatencyUtils's class to use is LatencyStats, which uses AtomicHistogram
> which uses Long to count, which means this has a bigger footprint than for
> example IntCountsHistogram (which uses Long for values , but int for counts
> which is enough in our context)
>
> Question 1 :
> - Does it look to you ok to stay on LatencyUtils ? for now we built our
> work on it
>
> Question 2:
> - Javadocs of LatencyStats#recordValue mentions nanosec, but it looks like
> we could use directly millis, can somebody confirm ?
>
>
> If some of you want to look at work in progress of PR:
>
> - https://github.com/ubikloadpack/jmeter/pull/50
>
>
> References:
> - https://groups.google.com/forum/#!topic/mechanical-sympathy/C6d-vFeYs_c
> - https://hdrhistogram.github.io/HdrHistogram/JavaDoc/org/
> HdrHistogram/AtomicHistogram.html
> - https://github.com/LatencyUtils/LatencyUtils/blob/
> d8f51f39f6146e1ad9a263dc916bcbc0ec06e16d/src/main/java/org/
> LatencyUtils/LatencyStats.java#L196
>
> Thanks
> Regards
> Qian Qian and Philippe
> --
>
>
> [image: logo Ubik Ingenierie] <https://www.ubik-ingenierie.com> Philippe
> Mouawad
> Senior Performance Expert
> 320914981 <+33320914981> | p.mouawad@ubik-ingenierie.com
> [image: ubik-ingenierie.com] ubik-ingenierie.com
> <https://www.ubik-ingenierie.com> | [image: 03.20.91.49.81] 03.20.91.49.81
> <+33320914981> | [image: 23 rue du chemin de fer , 59100 , Roubaix] 23
> rue du chemin de fer, 59100, Roubaix
> <https://www.openstreetmap.org/#map=18/50.69454/3.16455>
>
--
[image: logo Ubik Ingenierie] <https://www.ubik-ingenierie.com> Philippe
Mouawad
Senior Performance Expert
320914981 <+33320914981> | p.mouawad@ubik-ingenierie.com
[image: ubik-ingenierie.com] ubik-ingenierie.com
<https://www.ubik-ingenierie.com> | [image: 03.20.91.49.81] 03.20.91.49.81
<+33320914981> | [image: 23 rue du chemin de fer , 59100 , Roubaix] 23 rue
du chemin de fer, 59100, Roubaix
<https://www.openstreetmap.org/#map=18/50.69454/3.16455>