Parallel computing

[dxvn818]

Hello,

I would like to ask about the parallel computing. Normally, I only set the "ThreadNumber" in the first line of input file:

<GranOO Version="2.0" TotIteration="10000000" OutDir="..." Verbose="No" ThreadNumber="16">

However, the speed when "ThreadNumber=16" or "ThreadNumber=1" are not really different, so I think I must do something wrong in setting parallel computing in GranOO:





Do I need to set "ThreadNumber" in every plug-in ? Or what else should I do to make use of parallel computing ?

Thanks and regards,
Vinh

[Damien André]

Hello,
Multi-threading is only efficient if your simulation have domains higher than 10,000 discrete elements.
For "small" domains (with small number of discrete element), the benefice is null.

How many discrete element did you manage ?
Damien.

[dxvn818]

Hello Damien,

As I remember, I tested with 20000 discrete elements. Is it right that I only need to set "ThreadNumber" one time in the first line of input file ? Or must I set it in every plug-in ?

Could you please take a look at the input file where I tried setting multi-thread ?

PlasticTension.inp

Thank you very much,
Vinh

[jeremie.girardot]

Dear Vinh,

your .inp file is good.

Unfortunately, multithreading in GranOO is not set for all the available plugIns and is in general not very efficient at this moment. The GranOO team is working right now on this issue and will proposer soon a new more efficient version.

regards,

Jérémie Girardot

[dxvn818]

Hello Jeremie,

Thanks for your answer and looking forward to new update.

Regards,
Vinh

[Damien André]

Hello,
If you update granoo, you can see at the end of a computation the multi-threaded plug-in. They are highlighted with the '(MT)' acronym.

Damien.

[nnbbl]

Hello Damien,

I'm interested in your excellent work："Using the discrete element method to simulate brittle fracture in the indentation of a silica glass with a blunt indenter", and I have studied to simulate the model in this paper.  

In my test, the discrete silica sample is composed by 64000 discrete elements. It takes several hours to simulate even after MT calculation.

I wonder is there any solution to solve the time-consuming problem, or would you like to share the code in this paper and provide your simulation time.

Thank you very much.
Na

[Damien André]

Hello Na,
No, there is no way for accelerating the computation. MT is the sole solution. My advice is to start simulation campaign with low fineness around 2,000-5,000 discrete elements. It allows us to make quick trial and errors computations. It is required for debugging. When your parameters are okay, you can run "heavy" computations with 20,000-50,000 discrete elements.

For 50,000 discrete elements, it takes around 2 or 3 days of computation.
Kind regards, Damien.