Parallel Configuration

Submitted by symscape on May 30, 2013 - 16:42

Caedium's preferences control how the RANS Flow CFD solvers perform parallel simulations, also know as High Performance Computing (HPC). When configured correctly the parallel preferences will lead to significantly reduced simulation turnaround times on multi-core computers and clusters.

The parallel preferences are available when the RANS Flow or Professional Caedium add-ons are activated.

Parallel Preferences

To view the parallel preferences, start Caedium, select the File Toolbar, and click the Preferences button . In the Properties Panel open Physics->OpenFOAM->Parallel. The options for the Parallel preference are:

Shared Memory (default) - single computer with one or more cores
Cluster - multiple computers linked together in a cluster (For details see "Microsoft Windows HPC Server 2008 Cluster Configuration")

Both parallel options provide two preferences to configure parallel simulations:

Elements Per Process (default = 1000) - Minimum number of elements per process
Processes (default = maximum number of system processes x 1/2) - Maximum number of parallel processes

Parallel Preferences

Elements Per Process

The Elements Per Process preference provides a minimum cell limit to ensure there are enough cells for each parallel process to warrant using parallel. If there are too few cells the simulation may actually run slower than a serial (non-parallel) simulation. During parallel simulations there are inter-process communication overheads that for small simulations can negate the speed-up due to running multi-processes.

Processes

The Processes preference provides a maximum limit on the number of processes to use for parallel simulations, once the Elements Per Process condition is satisfied. If you want to perform other tasks while a parallel simulation is running, then consider reducing the Processes preference to free up a process - otherwise you are likely to experience sluggish response.

Comments

Increase element per process

Submitted by Sauzer on December 28, 2011 - 07:43.

Hi, I have a pc with this configuration:
Intel i7 - Ram 16 GB - Windows 7 professional 64 bit

I see that the cpu works at 12% and RAM at 14% during the simulations.
Do I need to set something in the software to use more cpu and ram?
(I followed this tip, the settings on my pc are these: Processes 7 - Element per process 10000;
I've increased element per process but i don't see changes).
Thank you

Reduce the Elements Per Process to utilize more cores

Submitted by symscape on December 28, 2011 - 10:01.

To increase the number of processors (cores) being used you need to decrease the Elements Per Process preference (try 1000). If you distribute a small case (low number of cells) among many processors the communication overhead can swamp any speedup resulting in a slow down.

Also you want the Processors preference = number of cores - 1. For your Intel i7 it appears to have 4 cores. However, on the Windows Task Manager it will show up as having twice as many cores (8) on the CPU Usage History monitor due to hyper threading. You are unlikely to see any significant speed up beyond the number of cores (4).

To get a feel for the speedup you can achieve try varying the two preferences with the Physics->OpenFOAM->Diagnostics = On - at the end of each solver run there will be an elapsed time printed in the File->Log.

Symscape