CFX not running in MPI distributed mode

evan247 · March 5, 2017, 07:19

Hi all,

I'm testing a CFX run on my university's HPC cluster using 2 nodes each with 16 CPU cores, hence a total of 32 processes. Having read relevant previous posts on setting up distributed parallel runs on this forum, I was able to start the simulation on both nodes, but soon it fails with information below from the .out file.

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

1. For the same simulation, I was able to run it in "MPI local parallel" mode on one of the nodes. So I don't think there's anything wrong with the CFD setup. Although I do experience unusual slowness in this local parallel run on a remote HPC node, compared to running it on my local machine.

2. My command is as below and if anyone can point out what I've done wrong, it'd be much appreciated.

cfx5solve -batch -def [filename].def -start-method "MPI Platform Distributed Parallel" -par-dist "[node list]"

Thank you.

evcelica · March 8, 2017, 16:40

Distributed requires setup of the distributed cluster.
Local parallel works out of the box.
It sounds like distributed wasn't set up properly on your university nodes.
Some of the setup is:
- Remote Desktop Acess
- hosts files and machine types populated
- MPI installed on all machines
- firewall exceptions
- Caching passwords if applicable
See the Installation documentation for more info.

March 5, 2017, 07:19	CFX not running in MPI distributed mode	#1
evan247 New Member Join Date: Jan 2015 Posts: 29 Rep Power: 11	Hi all, I'm testing a CFX run on my university's HPC cluster using 2 nodes each with 16 CPU cores, hence a total of 32 processes. Having read relevant previous posts on setting up distributed parallel runs on this forum, I was able to start the simulation on both nodes, but soon it fails with information below from the .out file. +--------------------------------------------------------------------+ \| An error has occurred in cfx5solve: \| \| \| \| The ANSYS CFX solver could not be started, or exited with return \| \| code 255. No results file has been created. \| +--------------------------------------------------------------------+ 1. For the same simulation, I was able to run it in "MPI local parallel" mode on one of the nodes. So I don't think there's anything wrong with the CFD setup. Although I do experience unusual slowness in this local parallel run on a remote HPC node, compared to running it on my local machine. 2. My command is as below and if anyone can point out what I've done wrong, it'd be much appreciated. cfx5solve -batch -def [filename].def -start-method "MPI Platform Distributed Parallel" -par-dist "[node list]" Thank you.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
CFX parallel hp MPI problem	fluidmechanics	CFX	5	June 19, 2013 19:05
Sgimpi	pere	OpenFOAM	27	September 24, 2011 07:57
HP MPI warning...Distributed parallel processing	Peter	CFX	10	May 14, 2011 06:17
Statically Compiling OpenFOAM Issues	herzfeldd	OpenFOAM Installation	21	January 6, 2009 09:38
Is Testsuite on the way or not	lakeat	OpenFOAM Installation	6	April 28, 2008 11:12

March 8, 2017, 16:40		#2
evcelica Senior Member Erik Join Date: Feb 2011 Location: Earth (Land portion) Posts: 1,184 Rep Power: 23	Distributed requires setup of the distributed cluster. Local parallel works out of the box. It sounds like distributed wasn't set up properly on your university nodes. Some of the setup is: - Remote Desktop Acess - hosts files and machine types populated - MPI installed on all machines - firewall exceptions - Caching passwords if applicable See the Installation documentation for more info.