Learning Forum

[Maarten Vanierschot]

I have a transient simulation using URANS running in parallel mode on 72 cores. However, I have noticed that as the simulation proceeds, the time per iteration gets longer and longer. If I run 3000 timesteps, the last timestep takes about 5 times longer than the first timestep. Could you help me with this issue as the simulations are very inefficient in this way...

[Nilay Pedram]

Hello,
One of the reason could be due to very less DOF of the simulation. Other reason could be the hardware so it is recommended to ensure that it is supported by checking the Ansys platform support while performing transient simulations.
Please go through the Fluent User's guide for more information on Parallel Processing - 42.1. Introduction to Parallel Processing (ansys.com)

Hope this helps you. Thanks!

[Maarten Vanierschot]

Thank you very much for your fast response! We have observed this behavior on hardware that should be supported (Intel IceLake 8360Y CPUs; a RHEL8-derived OS; with both ANSYS Fluent versions 2021R2 and 2022R2). Please also note that:

- it is a simple URANS simulation with only fluid flow and no special features
- these simulations employ dual-time stepping,
- the gradual slowdown also occurs when running on a single core (just a single MPI process),
- by printing timestamps at every step and iteration, we see that it is the individual iterations that are gradually taking more and more time,
- partial filling of RAM with file buffer caches can be excluded as a possible cause (the problem persists after flushing the cache).