Is my simulation time normal?

olli191 · August 11, 2015, 08:39

Hello CFD Community,

due to the fact that i am very new to CFD i have some questions concerning the calculation time.

Currently i am testing OpenFOAM on a Laptop with 8GB Ram and an Intel i5-5300 CPU at 2.3 GHz. Right now a Simulation with 1.3 Million Cells is running and will run for the next 7 days. The Simulation is transient RANS Flow of incompressible Air in a 2x7m room with an important feature of 1cm x 180cm (air inlet) and the solver is pimple. My System is Ubuntu 14.04 running with 4 Threads (2 Processors and 2 Hyperthreads).

Is the calculation time realistic or did i something really stupid? I know that the Hyperthreads are not helping but do they hinder a faster calculation?

Thanks for your help in advance

blspjnl · August 25, 2015, 15:50

Depends on loads of factors of course, but I'd expect 1 day rather than 1 week. (assuming 5000 iterations/day). Try a better initial guess next time to speed up convergence.

olli191 · September 8, 2015, 07:43

thanks for the reply

after some time i figured out, that the calculation time is like this due to the temporal resolution. After this idea i switched to a steady state simulation. As you see i did something stupid regarding my simulation problem.

Now to my next question:

How useful is a SSD for a CFD Simulation? Let me explain my question before you start rolling your eyes.

Does it make sense to have a big swap partition on a real system or is the computer system inappropriate for the simulation as soon as it uses the swap partition?

flotus1 · September 8, 2015, 09:59

A fast SSD can be useful if you write lots of data to disk during the simulation. For example transient result files.
But since even the fastest SSDs are orders of magnitude slower than RAM especially in terms of latency, using a SSD as swap space if you are running out of memory wont help. There is no substitute for enough RAM.
Latency Numbers Every Programmer Should Know

blspjnl · September 8, 2015, 15:44

Check the Task manager/Resource Monitor and see if your PC does not use its pagefile on the SSD/HDD, while running the simulation. (To ensure you have enough RAM.)

The time required to write a 'results.dat' file is usually significant less than the simulation time, SSD/HDD won't matter that much. Expect 10s-1min for a 500MB file. Don't write/store too many results of course, you'll find a practical optimum. (Like once every 2 hours with 3 saves?)

-Maxim- · September 9, 2015, 09:01

I'm using Ansys CFX and my support guy said that I should deactivate Hyper-Threading (in BIOS). Since the cores work at 100%, there's 'no room' for virtual cores. So the simulation won't be faster.
In my case, I also only need 8 HPC licenses instead of 16 - but you don't have that problem with OpenFOAM

There's also more overhead when you split the job in more mini-jobs. The splitting up and putting the parts together takes time as well...

August 11, 2015, 08:39	Is my simulation time normal?	#1
olli191 New Member Join Date: Aug 2015 Posts: 5 Rep Power: 10	Hello CFD Community, due to the fact that i am very new to CFD i have some questions concerning the calculation time. Currently i am testing OpenFOAM on a Laptop with 8GB Ram and an Intel i5-5300 CPU at 2.3 GHz. Right now a Simulation with 1.3 Million Cells is running and will run for the next 7 days. The Simulation is transient RANS Flow of incompressible Air in a 2x7m room with an important feature of 1cm x 180cm (air inlet) and the solver is pimple. My System is Ubuntu 14.04 running with 4 Threads (2 Processors and 2 Hyperthreads). Is the calculation time realistic or did i something really stupid? I know that the Hyperthreads are not helping but do they hinder a faster calculation? Thanks for your help in advance

September 8, 2015, 09:59		#4
flotus1 Super Moderator Alex Join Date: Jun 2012 Location: Germany Posts: 3,400 Rep Power: 47	A fast SSD can be useful if you write lots of data to disk during the simulation. For example transient result files. But since even the fastest SSDs are orders of magnitude slower than RAM especially in terms of latency, using a SSD as swap space if you are running out of memory wont help. There is no substitute for enough RAM. Latency Numbers Every Programmer Should Know olli191 likes this.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
time step directories naming issue	Andrea_85	OpenFOAM	3	April 3, 2014 08:38
mixerVesselAMI2D's mass is not balancing	sharonyue	OpenFOAM Running, Solving & CFD	6	June 10, 2013 09:34
plot over time	fferroni	OpenFOAM Post-Processing	7	June 8, 2012 07:56
Orifice Plate with a fully developed flow - Problems with convergence	jonmec	OpenFOAM Running, Solving & CFD	3	July 28, 2011 05:24
Time step in transient simulation	shib	FLUENT	0	June 17, 2010 13:07

August 25, 2015, 15:50		#2
blspjnl New Member Join Date: Aug 2015 Posts: 3 Rep Power: 10	Depends on loads of factors of course, but I'd expect 1 day rather than 1 week. (assuming 5000 iterations/day). Try a better initial guess next time to speed up convergence.

September 8, 2015, 07:43		#3
olli191 New Member Join Date: Aug 2015 Posts: 5 Rep Power: 10	thanks for the reply after some time i figured out, that the calculation time is like this due to the temporal resolution. After this idea i switched to a steady state simulation. As you see i did something stupid regarding my simulation problem. Now to my next question: How useful is a SSD for a CFD Simulation? Let me explain my question before you start rolling your eyes. Does it make sense to have a big swap partition on a real system or is the computer system inappropriate for the simulation as soon as it uses the swap partition?

September 8, 2015, 15:44		#5
blspjnl New Member Join Date: Aug 2015 Posts: 3 Rep Power: 10	Check the Task manager/Resource Monitor and see if your PC does not use its pagefile on the SSD/HDD, while running the simulation. (To ensure you have enough RAM.) The time required to write a 'results.dat' file is usually significant less than the simulation time, SSD/HDD won't matter that much. Expect 10s-1min for a 500MB file. Don't write/store too many results of course, you'll find a practical optimum. (Like once every 2 hours with 3 saves?)

September 9, 2015, 09:01		#6
-Maxim- Senior Member Maxim Join Date: Aug 2015 Location: Germany Posts: 415 Rep Power: 12	I'm using Ansys CFX and my support guy said that I should deactivate Hyper-Threading (in BIOS). Since the cores work at 100%, there's 'no room' for virtual cores. So the simulation won't be faster. In my case, I also only need 8 HPC licenses instead of 16 - but you don't have that problem with OpenFOAM There's also more overhead when you split the job in more mini-jobs. The splitting up and putting the parts together takes time as well...