How to setup CFX distributed which is compatible with Microsoft job scheduler in Windows HPC 2008? Pls help me I'm struggling do do the setup

[MadsR]

I think I solved this, if this still is an issue I may help you out. I know this is an old thread, but I just hate unsolved threads :-)

However, I can only start jobs from within cfx5solve and not from the HPC queue system. This gives me trouble when I want to do parametric runs - anyone familiar with this?

/Mads

[Attesz]

On Windows HPC is the simulation faster? Is it worth to change from XP? Thanks

[MadsR]

Hi Attilia - I have no idea. I prefer Linux and I am only using Windows HPC as it is mandatory to use Windows in our company (but I am trying to change that) - so I haven't ever tried to run CFD under XP.

/Mads

[Attesz]

Okay, thanks!

[MadsR]

Hi

I guess I figured this out. Not sure if this is the most streamlined method though, but it works beautifully.

I setup a job with tasks that perform the following command:

"\\<myHPCserver>\<myAnsysInstallDir>\ANSYS Inc\v121\CFX\bin"\cfx5solve -def <myDefinitionFile>.def -part 8 -double -par-local

to run a job with 8 threads. Windows HPC automatically distributes the threads on more machines, if I use more threads than is available on one machine.

If I want to run a parametric sweep (say for multiple angles of attack for an airfoil) I just put in a task for each angle of attach (eventhough I know there is a parametric sweep-job-option) and set up dependencies that each task must finish before the other. Quite easy actually. The whole thing is saved to a job file for reuse.

NOTES:
1. remember to install CFX into a path formulated as (UNC): \\<server>\<path>\ if you didn't do this, change all ANSYS-related paths in the environment variables to a UNC path.
2. remember the "-signs in the command above
3. you can put a cfx5pre -batch <mysession.pre> as the first task to setup your job if you make a loop in <mysession.pre> you can setup .def files for a parametric sweep.

More tips & hints are greatly appreciated :-)

/Mads