[hermedo]

Hi,

I'm simulating Fluid-Structure Interaction problem by system coupling between 'Transient Structural' and 'Fluent' modules on ANSYS 2019 R1 workbench.
I allocated 3 nodes (each has 32 cores CPU, 128GB memory capacity) on Fluent and 1 node (32 cores CPU, 128GB memory capacity) on Structural (totally 128 cores) as a computation nodes.
All of the computation nodes have the same specifications such as CPU and memory H/W model.
Also they (computation nodes) have been connected by a graphic node (16 cores CPU, 192GB memory capacity) and a master node with VNC connection.

However...
When I'm monitoring the analysis, the (graphic server) memory usage on the analysis was about 13% at the start point of the analysis.
Ok, that's fine.
But after about 7 seconds (calculation time) in the simulation, then the total memory usage has been shown up to about 60%.
Finally about 5 seconds (calculation time) later (at 12-second calculation point) in the simulation, the analysis has been terminated abnormally.
The error message, that I saw, was shown just like Linux operating system killed the process automatically.
I think it is because of the out-of-memory error. I heard that Linux os system can force to shut down the process if the memory allocation quantity of a single process exceeds the total amount of memory.
However, I don't know the exact reason why this happened.
I can't understand what's going on in my simulation...

So, I want to tell you all about my wonders as follows:

1. Is there any issues about memory leak problems on FSI analysis with System Coupling module (Transient Structural - Fluent) on ANSYS workbench?
- I can't easily understand why the amount of memory usage gets larger and larger with the calculation. The software can allocate larger memory because of the parameter variance. However, I think too much memory are using in my simulation.

2. Is there any methods for optimizing data I/O on parallel analysis?
- I think that the time spent for saving analysis results, especially Fluent .cas and .dat files, is too much. Certainly, at the analysis start point, saving result files do not spend much time (it is really continuous). However, it spends too much time when the analysis is going on mid-points.
- For example, on my FSI simulation (analysis set-up is shown below), it reaches at the 14.0sec calculation point to make FSI restart point on 6th July and the result saving process is continuing even until now. In other word, the data saving process (this single process) spends over 17 days...! The only .dat file (Fluent result file) has not been generated until now...
< FSI analysis set-up >
Fluent: Using LES viscous model (Cs=0.1), About 7.0M cells (tetrahedrons)
Transient Structural: Undamped, About 120K elements (tetrahedrons)
System Coupling: Interaction between Fluent and Transient Structural

If you need more information about my simulation case, then please make any mentions below.
Truly thanks.