Error when compute pair/gran/local with capillary models

Submitted by giuraso on
Attachment Size
error message 163.14 KB
testlocalcapillary.txt 3.81 KB
ckloss

Hi Guiseppe,

there is a known issue with compute pair/gran/local for such cases which will be fixed in the next release!

Best wishes
Christoph

Mon, 8 Feb 2016 - 13:08 Permalink
ckloss

Hi Guiseppe,

the next release will contain an update for that. Please check if it works for you!

best wishes
Christoph

Fri, 13 May 2016 - 13:59 Permalink

Hi,
(SEE EDIT)

*****UPDATE****
- I have this simulation with a capillary model for cohesion.
- The simulation run without any problem on 32 cores.
- If I add a compute pair/gran/local, a Segmentation Fault happens for 32 cores depending on the mapping of the processors!
- No ERROR for 8, 16 cores or single core.
- If I change the maxSeparationDistanceRatio to 2.0 I have the error for single core!

Do you think there is a bug?
********

*****ORIGINAL POST****

I am using a capillary model in my simulation, easo/capillary/viscous.
I need to dump local information, let's say contact forces, but there are some problem with the compute pair/gran/local...
The problem occurs when there is a long range force on particles (i.e. there is a capillary force but they are not in contact - no overlap)

For example if I do something like

compute cpp all pair/gran/local

run 1

compute c2 all reduce sum c_cpp[1]
variable K equal c_c2
thermo_style v_K

The simulation will run for a few timesteps... probably because I insert a loose packing and at the beginning we don't have contacts or liquid bridges... when particles begin to touch each other I still have no problem... then, when some particles in the system have liquid bridges (force but no overlap) I get the following error:

Signal: Segmentation fault (11)
Signal code: Address not mapped (1)
Failing at address: (nil)

see picture for the full message.

Does anyone has a solution or any hints, suggestions, directions?

Thank you!

Giuseppe

*********
:::EDIT::::

I added a script that can be run in a couple of minutes and gives the error I mentioned.

:::EDIT:::

I realized the script was running in two minutes on 32 cores... :D I uploaded a new script.

I reduced the number of particles... tried to run it on single core... AND... NO Segmentation Fault error!

I ran it again on 32 cores... Segmentation Fault error AGAIN!

I ran it again on 8 cores... NO Segmentation Fault error!
I ran it again on 16 cores... NO Segmentation Fault error!

I think this is something related to the neighbor search... I tried to increase the maxSeparationDistanceRatio, but on 32 cores I have the same problem. If I reduce the surfaceLiquidContent to 0... NO ERROR.

I ran the simulation setting up different mapping of the processors:

processors 4 2 4 - NO ERROR (50000 timesteps)

processors 2 4 4 - ERROR at timestep 3620
processors 4 4 2 - ERROR a timestep 420

-OK! NOW- try to change the maxSeparationDistanceRatio to 2.0 and the Segmentation Fault happens at 29330 for single core!

Any idea of what's going on?