Learning Forum

[jms26]

Hi,

I have been running into numerous issues while trying to run a conjugate heat transfer simulation in Fluent. My first issue I made a post regarding here: https://forum.ansys.com/discussion/38786/fluent-conjugate-ht-retain-cell-residuals-sigsegv-error#latest (Summary: turning on retain-cell-residuals and re-iterating causes a SIGSEGV error ONLY when the solid cell zone is activated, no clue why - attempts at resolution are listed in the post).

I was unable to resolve this, and ultimately resorted to deactivating the cell zone, running an isothermal simulation, and viewing high residual cells with ease. Now, I would like to re-activate the solid zone, and run a transient simulation to see how the heat flux into the solid domain changes the solid temperature over time. However, when I reactivate the solid zone: a very similar (or potentially same) error occurs. The first lines of the error are below:

==============================================================================

Node 1058: Process 273816: Received signal SIGSEGV.

==============================================================================

==============================================================================

Node 1064: Process 273822: Received signal SIGSEGV.

==============================================================================

*** Error in `/home1/apps/ANSYS/2021R2/v212/fluent/fluent21.2.0/linmic/3ddp_node/fluent_mpi.21.2.0': free(): corrupted unsorted chunks: 0x0000000016360ef0 ***

*** Error in `/home1/apps/ANSYS/2021R2/v212/fluent/fluent21.2.0/linmic/3ddp_node/fluent_mpi.21.2.0': free(): corrupted unsorted chunks: 0x00000000153faf50 ***

Along with this, error log files for seemingly every single process are outputted. I just discovered that when I run a parallel check in the Fluent GUI, I get the warning message below both when the solid zone is activated and deactivated:

"WARNING! Memory utilization is very high or imbalanced

WARNING! Load is imbalanced"

My only guess is that the issues I have been dealing with are related to this somehow. I use all the defaults for partitioning: Metis and have verified that solid zone weighting is activated. Is there anything I can do to aid in balancing the load? All machines that I am connecting to have the same number of processes, so I am unsure why the load is imbalanced. I am using over 1000 processes for a mesh around 30 million cells.

[Rob]

Can you try with 30 cores? I wonder if the partitions have got confused, 100k cells per core is a sensible minimum for optimal use of compute. With too many cores the overheads for partition communication can outweigh the benefit of more cpu.

[jms26]

Hi Rob Thank you for the reply - I just tried that and unfortunately with no luck. The parallel check still gave the same warning, and when I attempted to run it with the activated solid zone it still crashed with a SIGSEGV error. I also have completed a strong-scaling study and noted that the speed-up hadn't peaked yet using 1344 cores. Please let me know if there is anything else I should try, and thank you so much for the help.

[Rob]

Working here. Can you initialise the model and run an iteration after turning on the residual retention? That's with 22R1.

[jms26]

Hi Rob
Unfortunately that does not work, I am using 21R2. Do you have any other suggestions? Not sure if it makes a difference but I imported the mesh into Fluent from Pointwise; it does pass the mesh check and Fluent creates a shadow wall BC correctly when I activate the solid zone.

Also not sure how relevant this is but another issue I ran into while running the steady simulation with the deactivated solid zone is that Fluent is saving the case file every time it autosaves despite the case file seemingly not changing. In the TUI, I set it to auto-save only "if-mesh-is-modified" and it is still auto-saving for some reason despite the mesh definitely not changing (no dynamic meshes at all and no auto-adaptation occurring). Very strangely, if I use data obtained recently on the first ever initial case file with the exact same mesh and then run, the residuals instantly blow up - it only doesn't do this if I continue to use the newest case file. I have no clue what Fluent could be doing on the backend that's causing this.
Thanks a ton.

[Rob]

What is the mesh quality like? Check orth-quality (Mesh Checks) but also skew via the contours panel (turn off node values & compute).
If a model is failing from a run case it tends to suggest something weird is going on. If you read in the case & data again plot the velocity contour: is it sensible or "chequer board"?