Issues running simulations in parallel version

Discussion of numerical issues which are not directly linked to specific applications
Post Reply
jmueller
Posts: 11
Joined: Sat Apr 21, 2018 12:02 am
anti_bot: 333

Issues running simulations in parallel version

Post by jmueller » Fri Mar 20, 2020 3:37 pm

Hello,

Recently I've started to run simulations on the parallel version of MICRESS in hopes to speed up the diffusion calculations and reduce overall simulation time. I am able to run and complete simulations, but they are not completing as expected.

The first issue I have is that I apparently cannot have the molar volumes of phases read from the database. This is important for my simulations as there is a temporal gradient in temperature. The 'database' command is not accepted when I try to run simulations in parallel. There doesn't appear to be any other issues with coupling the database, just the molar volumes.

The second issue I'm having is that I do not get any nucleation during my simulations when they are ran in parallel. When I run the simulation on my desktop computer the nucleation that I input seems to work just fine, but on the parallel version the simulation appears not to even try to nucleate new grains. The simulations complete in the parallel version, but there is not much change from the initial condition other than grain growth/smoothing of boundaries.

Thanks,
Josh

ralph
Posts: 167
Joined: Wed Apr 27, 2011 4:42 pm
anti_bot: 333

Re: Issues running simulations in parallel version

Post by ralph » Fri Mar 20, 2020 5:23 pm

Never seen something like this.

Which MICRESS version do you use?
Can you share the input files? At least with support@micress.de

Best,
Ralph

ralph
Posts: 167
Joined: Wed Apr 27, 2011 4:42 pm
anti_bot: 333

Re: Issues running simulations in parallel version

Post by ralph » Mon Mar 23, 2020 1:38 pm

Josh,

reading molar volumes from database is possible with MICRESS 6.4 and above. You are using 6.2 in parallel.

For the first 400 s, I see no differences between serial and parallel version. So far, so good, at least in comparing versions.
Nucleation:
Bernd gave me the hint that setting the number of new nuclei to -1 might not be valid for version 6.2.

Did you really use version 6.2 on your desktop?

Ralph

jmueller
Posts: 11
Joined: Sat Apr 21, 2018 12:02 am
anti_bot: 333

Re: Issues running simulations in parallel version

Post by jmueller » Mon Mar 23, 2020 2:45 pm

Thanks for looking into this, Ralph.

I use 6.4 on my desktop. I'm not sure why we have 6.2 for our parallel version. We just had our license updated so that we could run a linux version of MICRESS in parallel on a supercomputer. Is it possible we received the wrong version when we updated our license?

Thanks,
Josh

ralph
Posts: 167
Joined: Wed Apr 27, 2011 4:42 pm
anti_bot: 333

Re: Issues running simulations in parallel version

Post by ralph » Mon Mar 23, 2020 3:27 pm

Don't know.
I will send you the download details for 6.4.

Ralph

ralph
Posts: 167
Joined: Wed Apr 27, 2011 4:42 pm
anti_bot: 333

Re: Issues running simulations in parallel version

Post by ralph » Mon Mar 23, 2020 3:35 pm

The license can be updated independent of the installation.
I guess only the license was updated.

Ralph

ralph
Posts: 167
Joined: Wed Apr 27, 2011 4:42 pm
anti_bot: 333

Re: Issues running simulations in parallel version

Post by ralph » Mon Mar 23, 2020 3:40 pm

Please have also a look at your sequential TabP.
When I activated nucleation in 6.2, most of the time was spent in TQ. This is a serial part and running it in parallel will give you no benefit.
Parallel is good for diffusion, stress, etc.

Ralph

jmueller
Posts: 11
Joined: Sat Apr 21, 2018 12:02 am
anti_bot: 333

Re: Issues running simulations in parallel version

Post by jmueller » Mon Mar 23, 2020 4:57 pm

Thanks, Ralph.

I'll take a look into getting 6.4 installed on our parallel machine.

I'm intending to run simulations with simultaneous interstitial and substitutional solute partitioning. On my desktop, these these types of simulations would always slow down substantially after the nucleation of the second phase once partitioning began. I'm hoping that the parallel version will speed up these diffusion calculations.

Thanks,
Josh

Bernd
Posts: 1504
Joined: Mon Jun 23, 2008 9:29 pm

Re: Issues running simulations in parallel version

Post by Bernd » Mon Mar 23, 2020 10:29 pm

Dear Josh,

Whether parallel computing will help or not can be judged from the .TabP output:
- If most time is spent in the diffusion solver (or stress, temperature and fluid flow solver if activated), then parallelization can help tremendously.
- If most time is spent in the interface treatment (PF-solver, list), then time-stepping is critical and can often been improved.
- If most time is spent in TQ, then updating intervals and relinearisation schemes must be optimized.

Bernd

Post Reply