Unreproductible simulation (one core, one computer, one script) with a tutorial example

[Clerc Adriane]

In the framework of my PhD thesis, I use Yade and shear numerical simulations.
While using a script based on the periodic simple shear tutorial example given on the web side, I noticed that my simulation is not reproducible : the physical time, the controlling and the output parameters differ when a sequential yade calculation is run several times on the same core, same computer, same day, same script, same sample.
I supposed there is/are several parameter(s) which is/are not initialized when closing/opening Yade. Do you know them ?

I use Yade program on a Ubuntu 18.

Thanks in advance

[Robert Caulk (rcaulk)]

Please provide all necessary information to allow a developer to reproduce your bug. For example, a running script, your method of yade installation, and exact results that you are observing from said script.

[Robert Caulk (rcaulk)]

makeCloud is a random generation of spheres [1]. Use the seed argument if you want the same packing twice.

[1]https://yade-dem.org/doc/yade.pack.html

[Clerc Adriane (clerca)]

Sorry I forgot to underline the fact that my packing is the same for each run. I create it once and I load it 4 times, closing and opening Yade between two runs.

[Robert Caulk (rcaulk)]

I cannot find anywhere in your scripts where you are saving scenes/packings, loading scenes/packings or resetting the scene.

[Jan Stránský (honzik)]

Hello,

> Sorry I forgot to underline the fact that my packing is the same for each run.

please provide the exact script you are using. There is no packing loading in the current script..
Or the information how you load the packing..

thanks
Jan

[Luc Sibille (luc-sibille)]

Hello,

Could it be related to that:
https://<email address hidden>/msg03562.html
even if you seem to use a single core?

Luc

Le 31/01/2019 à 10:47, Clerc Adriane a écrit :
> ** Attachment added: "Report"
> https://bugs.launchpad.net/yade/+bug/1814052/+attachment/5234557/+files/report_v2.pdf
>

--
Luc Sibille
Université Grenoble Alpes / IUT1 de Grenoble
Laboratoire 3SR: Sols, Solides, Structures, Risques

Tel lab.: [phone]
Tel IUT: [phone]

[Clerc Adriane (clerca)]

Robert Caulk (rcaulk)and Jan Stránský (honzik), from the yade interface the simulation can be saved and loaded, that is how I saved the initial packing and I loaded it each time.

Luc Sibille (luc-sibille) by one core I mean Yade is not running in parallel, so differences due to rounding in calculation should not exist, do they ?

[Bruno Chareyre (bruno-chareyre)]

>not running in parallel, so differences due to rounding in calculation should not exist, do they ?

No.
That's why I closed, then re-opened the bug. :)
Let's see if someone can reproduce it.

Bruno

[Luc Sibille (luc-sibille)]

> Luc Sibille (luc-sibille) by one core I mean Yade is not running in
> parallel, so differences due to rounding in calculation should not
> exist, do they ?
Actually I cannot answer, I do not know sufficiently computers and Yade
(not sure for instance there are not any part of the computation in
parallel, even if Yade is launched on a singe core ...). But this point
has been raised several times. So I guess that if your issue is related
to this point (and it is possible as your results are not fundammentally
different) some persons reading this mail will be able to answer it.
Luc

--
Luc Sibille
Université Grenoble Alpes / IUT1 de Grenoble
Laboratoire 3SR: Sols, Solides, Structures, Risques

Tel lab.: [phone]
Tel IUT: [phone]

[Robert Caulk (rcaulk)]

I am able to reproduce the described behavior using both makeCloud(seed=1) and regularHexa.

[Robert Caulk (rcaulk)]

*In all cases I am executing yade -j1 mwe.py

[Clerc Adriane (clerca)]

I think I have found the parameter which was making my simulation unreproducible :
In the Pyrunner, the checker is working with "realtime"... I didn't know but this the CPU time which can vary in function of how my computer is busy.
By replacing "realtime" by "iterPeriod", for a same packing loaded initially, the simulation is reproducible.

[Clerc Adriane (clerca)]

here is the packing

[Bruno Chareyre (bruno-chareyre)]

Good finding (#14). Not a bug then, closing.