[Daniel_Khazaei]

I'm running fluent simulation on windows 10 but it seems that I'm facing memory leak problem....

I'm running 4m grid with the following setup:
- Transient, VOF, SIMPLE, second order discretization
- Turbulent: Realizable K-Epsilon with menter-lechner
- Dynamic mesh, spring method

Here is the spec of the system I'm using:
- Core i7 6700K
- 32GB DDR4 RAM

At the beginning of the simulation it only uses 50% of the memory. However after 1 day, it almost uses all of the memory and finally get terminated with low memory error

What's the problem? how much RAM a 4m grid domain needs for the aforementioned case setup?

Best wishes,
Regards

[Daniel_Khazaei]

well after watching the simulation, it seems the memory consumption grows by every auto case and data writing
why does it keep case and data in memory after writing it to HDD?

[LuckyTran]

Are you sure it's not because of your dynamic meshing?

[sina_mech]

I recently did a transient multiphase simulation with ~6M cells on a linux cluster of ~150 cores (means too much communications involved) with less than 20GB of ram! My point is that there is something wrong!

It must be with dynamic meshing. Create a monitor and write the mesh-size as a text file so you can have a record of mesh size after it gets terminated. So you can check when/why the mesh gets too fine!

[Daniel_Khazaei]

Originally Posted by LuckyTran

Are you sure it's not because of your dynamic meshing?

Originally Posted by sina_mech

I recently did a transient multiphase simulation with ~6M cells on a linux cluster of ~150 cores (means too much communications involved) with less than 20GB of ram! My point is that there is something wrong!

It must be with dynamic meshing. Create a monitor and write the mesh-size as a text file so you can have a record of mesh size after it gets terminated. So you can check when/why the mesh gets too fine!

thank you for the recommendations, I will check the dynamic mesh setup for the grid size after re-meshing and get back to you.

[Daniel_Khazaei]

Originally Posted by sina_mech

I recently did a transient multiphase simulation with ~6M cells on a linux cluster of ~150 cores (means too much communications involved) with less than 20GB of ram! My point is that there is something wrong!

It must be with dynamic meshing. Create a monitor and write the mesh-size as a text file so you can have a record of mesh size after it gets terminated. So you can check when/why the mesh gets too fine!

Again after about 1 day the simulation get terminated with low memory error! I have checked the number of cells after 1500 time steps and the cell count never goes beyond 4.1m cell...well this is expected as the re-meshing is just used to avoid bad quality elements...

[sina_mech]

Originally Posted by Daniel_Khazaei

Again after about 1 day the simulation get terminated with low memory error! I have checked the number of cells after 1500 time steps and the cell count never goes beyond 4.1m cell...well this is expected as the re-meshing is just used to avoid bad quality elements...

Well, this is weird then. 4.1M is nothing for 32GB of ram. I would also try to check if the license manager is working well.

How long did it take to get to 1500 time steps? Is this 4.1M the latest mesh size? My point is that a crazy mesh-size increase can happen in just few time steps. You need to know the cell number for every time step.

[LuckyTran]

You could also just turn off dynamic meshing altogether. Do you actually need a dynamic mesh in your problem?

[Daniel_Khazaei]

Originally Posted by sina_mech

Well, this is weird then. 4.1M is nothing for 32GB of ram. I would also try to check if the license manager is working well.

How long did it take to get to 1500 time steps? Is this 4.1M the latest mesh size? My point is that a crazy mesh-size increase can happen in just few time steps. You need to know the cell number for every time step.

Well I get your point about rapid increase in cell count in few time steps, however by monitoring memory usage I have noticed that memory consumption increased gradually every 250 time steps (auto case and data writing period)!
It starts with 47% memory usage and then gradually increases after every case and data writing. For example from 47% to 53%, 60%, 68%, ..., 100%!!!Everything is normal when I check the written case and data files, even their size are consistent with the initial case and data! The calculation speed is also consistent...

I have checked all of the written case files: in a few steps the cell count drops to 3.8m, however the memory usage still increased

Originally Posted by LuckyTran

You could also just turn off dynamic meshing altogether. Do you actually need a dynamic mesh in your problem?

well I'm simulating a ship hull and the ship motion through waves is important for me!

[Daniel_Khazaei]

Well it seems that ANSYS 17.1 has memory management problem on windows 10 and keeps case and data in memory even after writing it to HDD.
Asked my supervisor to upgrade ANSYS to version 17.2 and the problem is now gone...memory usage remains at 47-50% all the time