[Shamoon Jamshed]

Hi,
It may be due to lack of computational resources, but as you know higher management demands results not excuses so I want my results in no time.

I am running LES since 4 months and after 1 month it has simulated 1milli sec phenomenon. I am simulating cylinder of 2 m with gas flow and at high pressure of 30 bar. The time step size have to be kept small that is 1e-06 sec and it is due to instability at larger time step size.

I need a shortcut in using LES. Computer is z800 with 8 cores (hyper threading makes 16 cores) and 64 GB ram. I need to finish it in 2 months as it is job problem. I need to simulate up to 1 sec physical time.

Please suggest quick solution.

Regards,

[CeesH]

I'm sorry, but life has no shortcuts.

If you already know that larger timesteps cause instability, then you're stuck with it. How big is your mesh size; can you reduce on spatial discretization without missing out on the scales of turbulence you want to resolve? If there is no space there too, well, this is probably the simulation you have to live with.

However, 1000 timesteps in 1 month does sound steep. Can you optimize the mesh partitioning such that the communication load between cores is lower (Although on a shared-memory machine, I doubt this would be a large bottleneck)

[ghost82]

Hi,
I totally agree.
You can improve a bit your speed by disabling the hyperthreading in the bios.

[Shamoon Jamshed]

Originally Posted by CeesH

I'm sorry, but life has no shortcuts.

If you already know that larger timesteps cause instability, then you're stuck with it. How big is your mesh size; can you reduce on spatial discretization without missing out on the scales of turbulence you want to resolve? If there is no space there too, well, this is probably the simulation you have to live with.

However, 1000 timesteps in 1 month does sound steep. Can you optimize the mesh partitioning such that the communication load between cores is lower (Although on a shared-memory machine, I doubt this would be a large bottleneck)

Thanks Ceesh, I should have mentioned the mesh size before. It is 3 million cells. I am running on default 16 cores which are shared. I have enabled hyperthreading.
"1000 timesteps in 1 mnth sound steep" what do you mean? Are you appreciating it or its not something good progress?