CFX Parallel Run Error
 

I am trying to run a job on local parallel and I keep getting the following error. I can't find anything in the help topics about it and I am not sure what to do!

"The ANSYS CFX solver could not be started, or exited with return code 255: . No results file has been created."

in which version you receive this error?

Post more of the output to give some more hints. Error 255 is not very helpful unfortunately.

CFX Parallel doesn't start/Serial works fine
 

Hey Guys!

I have the same problem. I can run my job on serial, but every parallel attempt shows the following in the out-file:

+--------------------------------------------------------------------+
| |
| Solver |
| |
+--------------------------------------------------------------------+


+--------------------------------------------------------------------+
| An error has occurred in cfx5solve: |
| |
| The ANSYS CFX solver could not be started, or exited with return |
| code 255: No such file or directory. No results file has been |
| created. |
+--------------------------------------------------------------------+

End of solution stage.

This run of the ANSYS CFX Solver has finished.
------------------------------------------------------------------------
The complete outfile is attached.

My system:
Windows 7 64 bit
ANSYS 14.5.7 64 bit
Intel Core I5 3570K @ 3,4 Ghz
8 GB Ram

The partitioning seems to work well, anyhow it looks everything okay in the out-file. I tried to use 2, 3, and 4 Cores with MPI local parallel and MPI distributed parallel but nothing worked. I also followed the Installation and Licensing Guide to ensure everything is right installed. The project itself also seems to be okay (mesh/BC's etc), because it runs on my notebook (Core2Duo) in MPI local parallel very well.
I read that it might help to use another parallel run mode like PVM or MPICH2, but i dont know how to set them up. I can't choose them in the CFX Solver Manager GUI so i don't even know if they are installed. Do they come along with CFX?
My only available options are: serial/Platform MPI local parallel/Platform MPI distributed parallel.
I also ensured that the programms could pass the windows firewall and the connection to the licence server is also fine. I am also sure I have the right license for parallel computing, because it runs on at least two cores on my notebook and not even that works on my desktop pc.

I looked through this forum a while, but couldn't find a solution that helped .:(

If you need additional information to help me I'll try hard to deliver them!

Regards

Any ideas? Anybody?

When you installed CFX did you also install MPI? If not, start up the install program again and install MPI.

Yes, I also did install MPI. Could there be another explanation why I doesn't work?

It seems like an installation or configuration issue since you have it working on your notebook. Sorry I don't have many other suggestions. Are you using Intel MPI or Platform MPI? Whichever you are using, have you tried the opposite?

I think I'm using Platform MPI but I will check on that, thanks!

Hi Googlestuff,
PVM and MPICH2 libraries - I believe has to be installed separately. However, though it is useful, I don't think that will solve your problem, because the out file seems to indicate that the partitioning is successful.

You could try progressively increasing the solver memory allocation factor, to 1.2 or 1.5, or as high as you can go, based on the memory being used by the simulation. This worked in a similar situation I faced.

Hey,
I wanted to try the Intel MPI and installed it. But I can't choose it in the CFX Solver Manger Window. Do i have to implement it somehow to ansys?

shreyasr do you mean the memory allocation factor of the solver, the interpolator or the partitioner? Or even all of them?

I'm not sure why you wouldn't be able to select Intel MPI in the Solver Manager. For me it has just worked.

Based on your output, it seems like the partitioning succeeded. To me it doesn't seem like a memory problem since it works in serial. But you can try adjusting the memory allocation factor if you'd like.

I think the best thing you can do to figure this out is run a very small and simple case in parallel and see if that works (maybe grab a small case from the CFX examples directory). If not, you have narrowed it down to a problem with the parallel configuration.

I agree with Chris's suggestion. That would make things clear.

@Chris : I agree with you, it does not seem apparently a memory problem. To try recollect my actions when I faced this error code - I tried looking it up, and found no useful information. The exception was in this thread :http://www.cfd-online.com/Forums/cfx...e-255-cfx.html .. You can see, from the last post that the simulation has worked on another machine. Well, interesting enough - if I remember right - on increasing the solver partition factor, my simulation was running. It is probably worth a shot anyway, if there's no useful information ? In my case - I could not try the serial option due to mesh size limitations. It was a simulation that took up most of 32GB RAM, with partition factors at 1. It just about crawled along!

@Sebastian :You could increase all three, but the partitioner seems okay, I would increase the solver's first.

Hello

I have got exactly the same problem and do not success in solving it.
I have tried several cases, small and big. They are solved in serial but as soon as I turn to parallel, i am getting the same problem as above.

Does anyone got a solution on that?

Thanks

Hi,
It could be possible that when the mesh is being split up for partitioning, there are elements of unacceptable quality created at the splits. This would not be visible in the overall mesh quality metrics. I suggest you could try matching interface meshes, if any, better or perhaps try marginally improving the quality. You could also give an attempt at varying the coupled and independent domain partitioning in the solver partitioning options, assuming there's more than 1 domain in your problem. Hope something works out!

Hi,
It could be possible that when the mesh is being split up for partitioning, there are elements of unacceptable quality created at the splits. This would not be visible in the overall mesh quality metrics. I suggest you could try matching interface meshes, if any, better or perhaps try marginally improving the quality. You could also give an attempt at varying the coupled and independent domain partitioning in the solver partitioning options, assuming there's more than 1 domain in your problem.

Btw: are you talking about a local parallel run or a distributed run with other computers? For the latter, I believe the admin has to setup access via his password for communication between the computers. The procedure is available in the Ansys documentation. I have also seen the 255 code when a computing node is not networked properly or if the license manager is stuck, or incorrectly configured. So you might wanna take a look at these things as well. Hope something works out !