reactingFoam faster on less CPUs

sufjanst · January 11, 2019, 04:41

Hello,

I am running a LES combustion case with PaSR (22 species, 124 reactions) on OpenFOAM 6. I tried to run the simulation on 80 CPUs and 16 CPUs. I have discovered that the simulation with less CPUs was running way faster than the other one. The bottleneck is in the calculation of the comustion/chemistry.

Can anybody tell me why this is happening?

Greetings
sufjanst

Abishek · January 11, 2019, 12:00

1. how many physical cores does the machine have?

2. Have you carried out a scaling study by looking at the speed-up vs Ncores -- typically increasing one node at a time (not core)?

pete20r2 · January 12, 2019, 09:26

Are you renumbering the mesh? What is the interconnect?
What is the decomposition method?
How many cells?

January 11, 2019, 04:41	reactingFoam faster on less CPUs	#1
sufjanst Member Join Date: Mar 2016 Posts: 73 Rep Power: 10	Hello, I am running a LES combustion case with PaSR (22 species, 124 reactions) on OpenFOAM 6. I tried to run the simulation on 80 CPUs and 16 CPUs. I have discovered that the simulation with less CPUs was running way faster than the other one. The bottleneck is in the calculation of the comustion/chemistry. Can anybody tell me why this is happening? Greetings sufjanst

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
How to switch off combustion and reaction in reactingFoam	shenzhou1987	OpenFOAM Running, Solving & CFD	16	October 26, 2017 16:31
strange processor boundary behavior in foam-extend reactingFOAM	Neka	OpenFOAM Bugs	8	August 16, 2017 08:13
What do you all do to stabilize reactingFoam (or in general)?	KarenRei	OpenFOAM Running, Solving & CFD	7	December 11, 2016 15:34
Flamelet gives faster chemistry than ED-model	champshof	FLUENT	0	November 30, 2011 11:01
reactingFoam wedge handling wrong U	dhondupant	OpenFOAM Bugs	1	December 9, 2010 08:34

January 11, 2019, 12:00		#2
Abishek New Member Join Date: Sep 2011 Posts: 11 Rep Power: 15	1. how many physical cores does the machine have? 2. Have you carried out a scaling study by looking at the speed-up vs Ncores -- typically increasing one node at a time (not core)?

January 12, 2019, 09:26		#3
pete20r2 Senior Member Peter Baskovich Join Date: Jul 2014 Posts: 127 Rep Power: 12	Are you renumbering the mesh? What is the interconnect? What is the decomposition method? How many cells?