Learning Forum

[hbahrein]

Greetings,

I've been diligently developing a head-neck biomechanical model. I successfully resolved issues on my personal computer, and it's now operating smoothly. However, when I attempt to execute the very same model, using the identical key file, on our local high-performance computing center, I encounter a negative volume error right from the outset.

To date, my working hypothesis is that the problem may be attributed to the transition from SMP to MPP computing environments. However, my understanding was that the solver algorithm remains consistent between the two, with the only discrepancy being in resource allocation, such as RAM and CPU.

I would really appreciate it if you could help me resolve this issue.

Thanks,

[Reno Genest]

Hello Hossein,

There are differences in the SMP and MPP implementations of LS-DYNA and so running the same .k file with SMP might give slighthly different results compared with running MPP. The main difference between SMP and MPP is in the contact algorithm implementation.

I would recommend that you run MPP on your personal computer as well to see if you get the negative volume error. If so, you can then debug this problem.

Also, it is recommended to always run with the same number of MPP cores. This is because the domain of your model will be decomposed (or split) into n number of cores when running MPP. When using different number of cores, the decomposition will be different and the results will change slightly. If your model is close to instability, changing the number of MPP cores may be enough to get the negative volume error. The negative volume error is an indication of instabilities.

You will find more information about MPP here:

http://ftp.lstc.com/anonymous/outgoing/support/PRESENTATIONS/mpp_201305.pdf


Reno.

 

 

[hbahrein]

Thank you for your response. I am mostly using tied_surface_to_surface contacts with SOFT option 2 enabled.

Do you know what are the major differences between SMP an MPP regarding this contact formulation?

Knowing this will help me understand the problem better.

[Reno Genest]

Hello Hossein,

I don't know. We would have to ask a developer and compare the contact routines (SMP vs MPP).

If you intend to use MPP, I recommend you build, calibrate, and validate your model with MPP and stick to the same number of cores. For example, always use 32 cores with this model.

For the negative volume problem, you need to troubleshoot it with MPP. Please create a new post if you need help with negative volume error.

 

Reno.