Licensing

Licensing

  1. All Categories
  2. Additional Products - Licensing
  3. Mechanical APDL distributed solver 2020 R1 (may also apply for other versions) on Windows 10, may print extreme slow (in 10s MB/s or 200-300MB/s, should be in 4-5GB/s) “Communication speed from master to core #” in output file and slow down the solving: Communication speed from master to core 1 = 36.96 MB/sec Communication speed from master to core 2 = 36.05 MB/sec Communication speed from master to core 3 = 33.20 MB/sec Communication speed from master to core 4 = 44.90 MB/sec Communication speed from master to core 5 = 35.06 MB/sec Communication speed from master to core 6 = 32.52 MB/sec Communication speed from master to core 7 = 32.37 MB/sec Communication speed from master to core 8 = 35.84 MB/sec Communication speed from master to core 9 = 43.41 MB/sec Communication speed from master to core 10 = 38.41 MB/sec Communication speed from master to core 11 = 34.43 MB/sec Communication speed from master to core 12 = 44.37 MB/sec Communication speed from master to core 13 = 35.71 MB/sec Communication speed from master to core 14 = 36.46 MB/sec Communication speed from master to core 15 = 46.46 MB/sec Communication speed from master to core 16 = 41.15 MB/sec Communication speed from master to core 17 = 36.24 MB/sec Change to IBM MPI or MS MPI, same issue. Run MAPDL MPI test, everything works as expected: latency = 0.3398 microseconds Bytes Bandwidth(MB/s) ——- —————– 8 23.544842 1024 1613.448247 4096 3162.126055 16384 4319.923306 65536 2234.998517 262144 4293.039927 1048576 4850.385359 4194304 5141.109631 The /parallel/bandwidth In Fluent prints expected values (4-5GB/s). Only Mechanical APDL distributed solving gets such extreme slow MPI bandwidth issue.

Mechanical APDL distributed solver 2020 R1 (may also apply for other versions) on Windows 10, may print extreme slow (in 10s MB/s or 200-300MB/s, should be in 4-5GB/s) “Communication speed from master to core #” in output file and slow down the solving: Communication speed from master to core 1 = 36.96 MB/sec Communication speed from master to core 2 = 36.05 MB/sec Communication speed from master to core 3 = 33.20 MB/sec Communication speed from master to core 4 = 44.90 MB/sec Communication speed from master to core 5 = 35.06 MB/sec Communication speed from master to core 6 = 32.52 MB/sec Communication speed from master to core 7 = 32.37 MB/sec Communication speed from master to core 8 = 35.84 MB/sec Communication speed from master to core 9 = 43.41 MB/sec Communication speed from master to core 10 = 38.41 MB/sec Communication speed from master to core 11 = 34.43 MB/sec Communication speed from master to core 12 = 44.37 MB/sec Communication speed from master to core 13 = 35.71 MB/sec Communication speed from master to core 14 = 36.46 MB/sec Communication speed from master to core 15 = 46.46 MB/sec Communication speed from master to core 16 = 41.15 MB/sec Communication speed from master to core 17 = 36.24 MB/sec Change to IBM MPI or MS MPI, same issue. Run MAPDL MPI test, everything works as expected: latency = 0.3398 microseconds Bytes Bandwidth(MB/s) ——- —————– 8 23.544842 1024 1613.448247 4096 3162.126055 16384 4319.923306 65536 2234.998517 262144 4293.039927 1048576 4850.385359 4194304 5141.109631 The /parallel/bandwidth In Fluent prints expected values (4-5GB/s). Only Mechanical APDL distributed solving gets such extreme slow MPI bandwidth issue.

Tagged: , , , , , , ,

    • January 25, 2023 at 7:28 am
      FAQFAQ
      Participant

      Checked .dll used by ANSYS.exe in Process Explorer and found below suspected .dll file: umppc11009.dll CrowdStrike Falcon Sensor Support Module CrowdStrike, Inc. C:WindowsSystem32umppc11009.dll Renaming this file could resolve this issue: Communication speed from master to core 1 = 3664.57 MB/sec Communication speed from master to core 2 = 4403.35 MB/sec Communication speed from master to core 3 = 4252.71 MB/sec Communication speed from master to core 4 = 2462.62 MB/sec Communication speed from master to core 5 = 3032.25 MB/sec Communication speed from master to core 6 = 4411.23 MB/sec Communication speed from master to core 7 = 4429.14 MB/sec Communication speed from master to core 8 = 1096.84 MB/sec Communication speed from master to core 9 = 4244.59 MB/sec Communication speed from master to core 10 = 4403.03 MB/sec Communication speed from master to core 11 = 4393.25 MB/sec Communication speed from master to core 12 = 4377.22 MB/sec Communication speed from master to core 13 = 4321.73 MB/sec Communication speed from master to core 14 = 4430.78 MB/sec Communication speed from master to core 15 = 4381.27 MB/sec Communication speed from master to core 16 = 4387.57 MB/sec Communication speed from master to core 17 = 4299.54 MB/sec And solving performance is back to expected, 2-3X faster than before renaming. Assuming something in security software slows down the MPI communication among ANSYS.exe processes.

Featured Articles

Introducing Ansys Electronics Desktop on Ansys Cloud

The Watch & Learn video article provides an overview of cloud computing from Electronics Desktop and details the product licenses and subscriptions to ANSYS Cloud Service that are...

How to Create a Reflector for a Center High-Mounted Stop Lamp (CHMSL)

This video article demonstrates how to create a reflector for a center high-mounted stop lamp. Optical Part design in Ansys SPEOS enables the design and validation of multiple...

Introducing the GEKO Turbulence Model in Ansys Fluent

The GEKO (GEneralized K-Omega) turbulence model offers a flexible, robust, general-purpose approach to RANS turbulence modeling. Introducing 2 videos: Part 1 provides background information on the model and a...

Postprocessing on Ansys EnSight

This video demonstrates exporting data from Fluent in EnSight Case Gold format, and it reviews the basic postprocessing capabilities of EnSight.

© 2023 Copyright ANSYS, Inc. All rights reserved.