How much time does your code take to finish flux calculations?

aerosayan · May 20, 2021, 10:43

Trying to see what kind of performance others are observing.

Considering code like this:

Code:

do t = 1, ntimesteps
    do m = 1,4
        calculate_flux()
        calculate_residuals()
        update_solution()
    end
end

I'm specifically looking to see : How much time is required to finish *one call to the flux calculation function*, considering, your code solves 2D NS/Euler equations for a certain number of cells?"

If you don't have a 2D solver, but can guess how much time would be considered good for performance, please share it, as it will be helpful too.

If you have a 3D solver, please share the results too, as we will find it helpful.

Thanks

Eifoehn4 · May 23, 2021, 10:06

If you want to compare the performance of your code, i think you have to be more specific.

Mesh topology? Structured, unstructured?
Mesh metrics? Cartesian, non-Cartesian?
Cell type? Quad, Tria?
First, second, or higher order?
Numerical method? FV, FD, FEM, DG, WENO?
Does your flux calculation only considers the Riemann solver operation or other stuff, e.g. rotation in the facial coordinate system or reconstruction stuff?
What kind of flux function? Rusanov, HLLC, Roe?

Moreover, why do you want to normalize your results with the number of cells? The flux calculation is a facial operation.

Regards

Edit: Example

# Cores: 1 Core
Calculation time: ~20 seconds
Physical simulation time: 2 seconds
Equation system: Euler equations
Numerical method: FV
Flux function: Rusanov
Temporal order: Midpoint method O(2)
Spacial order: TVD MINMOD O(2)
Cartesian mesh
n_elems=1008
n_sides=2096

Note that the mesh in the picture does not represent the FV mesh, since my solver is based on a DG framework.

aerosayan · May 23, 2021, 11:01

Thanks for sharing your results.

Asking for results here by being very specific doesn't work. If I ask for the results from a very specific kind of solver design, maybe only the experts will be able to answer. Maybe not, as they may be only working with different methods.

It's best to ask for performance of their own solvers. I can figure out where my own code's performance needs to be by studying their results.

May 20, 2021, 10:43	How much time does your code take to finish flux calculations?	#1
aerosayan Senior Member Sayan Bhattacharjee Join Date: Mar 2020 Posts: 495 Rep Power: 8	Trying to see what kind of performance others are observing. Considering code like this: Code: do t = 1, ntimesteps do m = 1,4 calculate_flux() calculate_residuals() update_solution() end end *I'm specifically looking to see : How much time is required to finish one call to the flux calculation function, considering, your code solves 2D NS/Euler equations for a certain number of cells?"* If you don't have a 2D solver, but can guess how much time would be considered good for performance, please share it, as it will be helpful too. If you have a 3D solver, please share the results too, as we will find it helpful. Thanks

May 23, 2021, 10:06		#2
Eifoehn4 Senior Member - Join Date: Jul 2012 Location: Germany Posts: 184 Rep Power: 13	If you want to compare the performance of your code, i think you have to be more specific. Mesh topology? Structured, unstructured? Mesh metrics? Cartesian, non-Cartesian? Cell type? Quad, Tria? First, second, or higher order? Numerical method? FV, FD, FEM, DG, WENO? Does your flux calculation only considers the Riemann solver operation or other stuff, e.g. rotation in the facial coordinate system or reconstruction stuff? What kind of flux function? Rusanov, HLLC, Roe? Moreover, why do you want to normalize your results with the number of cells? The flux calculation is a facial operation. Regards Edit: Example # Cores: 1 Core Calculation time: ~20 seconds Physical simulation time: 2 seconds Equation system: Euler equations Numerical method: FV Flux function: Rusanov Temporal order: Midpoint method O(2) Spacial order: TVD MINMOD O(2) Cartesian mesh n_elems=1008 n_sides=2096 Note that the mesh in the picture does not represent the FV mesh, since my solver is based on a DG framework. aerosayan and aero_head like this. __________________ Check out my side project: A multiphysics discontinuous Galerkin framework: Youtube, Gitlab. Last edited by Eifoehn4; May 23, 2021 at 11:10.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
[Other] Contribution a new utility: refine wall layer mesh based on yPlus field	lakeat	OpenFOAM Community Contributions	58	December 23, 2021 02:36
AMI speed performance	danny123	OpenFOAM	21	October 24, 2020 04:13
Micro Scale Pore, icoFoam	gooya_kabir	OpenFOAM Running, Solving & CFD	2	November 2, 2013 13:58
plot over time	fferroni	OpenFOAM Post-Processing	7	June 8, 2012 07:56
Could anybody help me see this error and give help	liugx212	OpenFOAM Running, Solving & CFD	3	January 4, 2006 18:07

May 23, 2021, 11:01		#3
aerosayan Senior Member Sayan Bhattacharjee Join Date: Mar 2020 Posts: 495 Rep Power: 8	Thanks for sharing your results. Asking for results here by being very specific doesn't work. If I ask for the results from a very specific kind of solver design, maybe only the experts will be able to answer. Maybe not, as they may be only working with different methods. It's best to ask for performance of their own solvers. I can figure out where my own code's performance needs to be by studying their results.