[magjohan]

Hi Ansys RSM users,

I'm having issues to start CFX simulations using parallel computing. It works fine in serial mode but when trying to solve in parallel it does not work.

I have confirmed that Intel MPI are working when solving local and not by RSM.


I get these messages out from the RSM log:
Running Solver : C:\Program Files\ANSYS Inc\v161\CFX\bin\cfx5solve.exe -batch -ccl "runInput.ccl" -fullname "Fluid Flow CFX_021" -par-local -part 30 -start-method "Platform MPI Local Parallel"
[SOLVER STDERR] An error has occurred in cfx5solve:
[SOLVER STDERR] The ANSYS CFX solver could not be started, or exited with return code 255:
[SOLVER STDERR] . No results file has been created.
Command Exit Code: 2


How do I change it from using "platform MPI Local Parallel" to Intel?

[magjohan]

Originally Posted by magjohan

Hi Ansys RSM users,

I'm having issues to start CFX simulations using parallel computing. It works fine in serial mode but when trying to solve in parallel it does not work.

I have confirmed that Intel MPI are working when solving local and not by RSM.


I get these messages out from the RSM log:
Running Solver : C:\Program Files\ANSYS Inc\v161\CFX\bin\cfx5solve.exe -batch -ccl "runInput.ccl" -fullname "Fluid Flow CFX_021" -par-local -part 30 -start-method "Platform MPI Local Parallel"
[SOLVER STDERR] An error has occurred in cfx5solve:
[SOLVER STDERR] The ANSYS CFX solver could not be started, or exited with return code 255:
[SOLVER STDERR] . No results file has been created.
Command Exit Code: 2


How do I change it from using "platform MPI Local Parallel" to Intel?

Problem Solved!

###########################################
Part of the solution: "From ANSYS Customer Portal"
Knowledge Resources: Solutions

#2037824
Q: Platform MPI parallel runs of CFX or Fluent run slowly and produce the error message "hwloc has encountered what looks like an error from the operating system".

Product Family: Installation, Licensing & Systems
Product: ANSYS CFX
Version: 15.0
Area: HPC/Parallel
SubArea: N/A
Last Updated: Apr 28 2015


Answer:

The problem is that the hwloc error prevents Platform MPI from correctly binding the parallel processes to unique cores and so multiple processes can end up on the same core. This appears to be related to the new v2 intel chips.

Known workarounds are (in order of preference):
(1) Run Intel MPI. Intel MPI has binding automatically built into it.
(2) Disable hyper threading. This will require modifying the BIOS.
(3) In CFX the explicit CPU binding for HPMPI can be disabled by editing the start methods in C:\Program Files\ANSYS Inc\v150\CFX\etc\start-methods.ccl. In version 15.0 remove the "-cpu_bind" arguments. In version 16.0 remove the "-affcycle=socket -affwidth=numa" arguments. Running with no binding is still much slower than running with a correctly partitioned binding, but it should be much improved over an incorrectly partitioned binding.

Other CPU binding related solutions:
2037005
2031338
2037825

###########################################

Since Ansys RSM automatically uses Platform MPI when submitting jobs, I had to disable hyper threading and bam it turned out to solve the issue!