Fluids

Fluids

Topics related to Fluent, CFX, Turbogrid and more.

Fluent solution slowing down : 2-way FSI

    • sampathsrikanth1999
      Subscriber

      Hello everyone. I am trying to run a 2-way FSI in an artery. I am simulating using 8 cores in our University's cluster. The solution, in the beginning, was fast. With progress in time, I found that the simulation slowing down and the reason appears to be the mesh update during the next time step. The mesh update and the data distribution to the fluent cores is taking so much time. Once it is done, the simulation runs quickly. Is there anything I could do to speed up the simulation?

    • Rahul
      Ansys Employee

      You may want to review your dynamic mesh settings or consider refining your mesh to accommodate the deformations better without requiring extensive updates. This could involve adjusting the mesh quality criteria or implementing local mesh refinement in areas of high deformation.

    • sampathsrikanth1999
      Subscriber

      Hi Rahul. I tried to implement what you said. I refined the mesh (Now I have 1 M cells running on 40 cores). I changed my dynamic mesh settings. Relaxed the remeshing settings. I have spring based smoothing (parameters set according to the reed-valve tutorial) and local face and local cell based remeshing. I also tried to activate implicit update (once in every 10 iterations with relaxation factor 0.5) but I don't see it getting reflected in the fluent transcript. The mesh update during the next time step is still taking too much time (approx 12 minutes). This update time is so slow when flow into the artery reaches the maximum. Do you have any other suggestions that I can try? Thanks in advance

    • Rahul
      Ansys Employee

      The solution might slow down with more number of cores. With increasing processors, the mesh and data will be partitioned and each core do its part of the work and interchange info (the compute node does this). So if the number of processors is high, communication time for the CPU will be more than analysis time which is not effective. Try using fewer cores for Fluent (10-15) as you have about 1M cells. 

      If it is Linux machine try clearing Linux File Cache buffers, you can include the -cflush option when launching Ansys Fluent from the command line to flush the cache buffers.

    • sampathsrikanth1999
      Subscriber

      Hi Rahul. I ran the case using 15 cores just as you said but I am still facing the same problem. The mesh update is so slow. 

      "COUPLING STEP = 111      COUPLING ITERATION = 1". It gets stuck at this place. Its been 30 minutes(still counting). I also tried running a coarser mesh (140k cell in 2 cores). It also encountered the same problem. With time it got slower (it took 5 hrs to calculate 10 coupling steps (step size 1ms)). In the coarser mesh, I also encountered a issue that fluent is not converging. The fluent solution going well till now but now continuity is oscillating around 5e-3 (target 1e-4). 

       

    • sampathsrikanth1999
      Subscriber

       

      A small update. I am confused with the output file.

      COUPLING STEP = 111      COUPLING ITERATION = 1
       
      Updating solution at time levels N and N-1.
       done.
       
      Updating mesh at time level N...
      >> Dividing domain into 15 partitions using Metis.
         Time = 0.0999391 seconds.
         Done.
       
      It says the time taken was was less than 0.1 seconds. It took nearly 40minutes to output this line. What does this mean?
Viewing 5 reply threads
  • You must be logged in to reply to this topic.