OpenFOAM Multi-Core Solving across Physical vs. Logical Cores in Virtual Machine

Vyssion · September 28, 2016, 10:53

So I have been looking for ways to optimise my CFD code to run faster and more stable, and have come across this conundrum that a mate of mine pointed out might not be the case...

According to a few sources, when solving across multiple cores, there is very little to no benefit in decomposing over more than the number of physical cores that you have in your machine; that is, hyper-threading offers no substantial benefit.

My issue with this is that for a native linux build, sure, make sense. But I am not running natively - but rather inside a Virtual Box VM...

So I am wondering whether the same applies to VM setups; especially since the Virtual Machine doesn't "see" the same thing that the native OS does and may not be able to see whether a core is physical or logical. I had assumed that VMs and Native builds functioned differently and at the moment, I run the VM across 6 cores reserving two for my native OS and then decompose OpenFOAM into 8 processor cores which solve. I need to change that come to think of it as I write this!!

But yeah, so should I decompose into a (2 2 1) or would there be some benefit to perhaps decompose into a (2 3 1) if the VM can't tell the difference between the physical and logical cores. Or leave it as (2 2 2).

anishtain4 · September 28, 2016, 12:34

The best way to say is to have test runs same as the last post in the topic you mentioned.

September 28, 2016, 10:53	OpenFOAM Multi-Core Solving across Physical vs. Logical Cores in Virtual Machine	#1
Vyssion New Member Vyssion Join Date: Mar 2016 Posts: 12 Rep Power: 10	So I have been looking for ways to optimise my CFD code to run faster and more stable, and have come across this conundrum that a mate of mine pointed out might not be the case... According to a few sources, when solving across multiple cores, there is very little to no benefit in decomposing over more than the number of physical cores that you have in your machine; that is, hyper-threading offers no substantial benefit. My issue with this is that for a native linux build, sure, make sense. But I am not running natively - but rather inside a Virtual Box VM... So I am wondering whether the same applies to VM setups; especially since the Virtual Machine doesn't "see" the same thing that the native OS does and may not be able to see whether a core is physical or logical. I had assumed that VMs and Native builds functioned differently and at the moment, I run the VM across 6 cores reserving two for my native OS and then decompose OpenFOAM into 8 processor cores which solve. I need to change that come to think of it as I write this!! But yeah, so should I decompose into a (2 2 1) or would there be some benefit to perhaps decompose into a (2 3 1) if the VM can't tell the difference between the physical and logical cores. Or leave it as (2 2 2). __________________ *Optimism, pessimism, fck that: We're going make it happen. As God is my bloody witness, I'm hell-bent on making it work.** - Elon Musk in 2008, after three unsuccessful Falcon launches.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
chtMultiRegionSimpleFoam: maximum number of iterations excedeed.	Nkl	OpenFOAM Running, Solving & CFD	19	October 10, 2019 02:42
conjugate heat transfer in OpenFOAM	skuznet	OpenFOAM Running, Solving & CFD	99	March 16, 2017 05:07
Floating point exception error	lpz_michele	OpenFOAM Running, Solving & CFD	53	October 19, 2015 02:50
Moving mesh	Niklas Wikstrom (Wikstrom)	OpenFOAM Running, Solving & CFD	122	June 15, 2014 06:20
Micro Scale Pore, icoFoam	gooya_kabir	OpenFOAM Running, Solving & CFD	2	November 2, 2013 13:58

September 28, 2016, 12:34		#2
anishtain4 Senior Member Mahdi Hosseinali Join Date: Apr 2009 Location: NB, Canada Posts: 273 Rep Power: 18	The best way to say is to have test runs same as the last post in the topic you mentioned.