Very slow fluid simulation

[PavelBlend]

Question: Why is the simulator so slow? On my system, I simulate 15,000 particles for a long time.
Using the FLIP Fluids simulator on my system, I can simulate millions of particles. But with the help of splish slpash it turns out to simulate tens of thousands of particles.
Maybe I'm not correctly specifying the simulation parameters?
Here are the parameters of my simulation:

{
    "Configuration":
    {
        "pause": true,
        "timeStepSize": 0.0005,
        "numberOfStepsPerRenderUpdate": 2,
        "particleRadius": 0.01, 
        "density0": 1000, 
        "simulationMethod": 4,		
        "gravitation": [0.1,-9.81,0], 
        "cflMethod": 0, 
        "cflFactor": 0.05,
        "cflMaxTimeStepSize": 0.005,
        "maxIterations": 100,
        "maxError": 0.0001,
        "maxIterationsV": 1000,
        "maxErrorV": 0.001,		
        "stiffness": 50000,
        "exponent": 7,
        "velocityUpdateMethod": 0,
        "enableDivergenceSolver": true,
        "boundaryHandlingMethod": 2,
        "enablePartioExport": true,
        "dataExportFPS": 60
    },

    "Fluid":
    {
        "density0": 1000, 
        "colorField": "velocity",
        "colorMapType": 1,
        "renderMinValue": 0.0,
        "renderMaxValue": 5.0,
        "surfaceTension": 1.0,
        "surfaceTensionMethod": 0,		
        "viscosity": 3000.0,
        "viscosityMethod": 7, 
        "viscosityBoundary": 3000.0,
        "vorticityMethod": 1,
        "vorticity": 0.15,
        "viscosityOmega": 0.05,
        "inertiaInverse": 0.5,
        "maxEmitterParticles": 100000,
        "emitterReuseParticles": false,
        "emitterBoxMin": [-4.0,-1.0,-4.0],
        "emitterBoxMax": [0.0,4,4.0]
    },

    "Emitters": [
        {
            "width": 14,
            "height": 2,
            "translation": [0.0, 0.0, 0.0],
            "rotationAxis": [0, 0, 1],
            "rotationAngle": -1.57,
            "velocity": 3,
            "emitStartTime": 0,
            "type": 0
        }
    ],

    "RigidBodies": [
        {
            "geometryFile": "collision.obj",
            "color": [0.1, 0.4, 0.6, 1.0], 
            "isDynamic": false,
            "isWall": true,
            "mapInvert": false, 
            "mapThickness": 0.0,
            "mapResolution": [20,20,20],
            "samplingMode": 1
        }
    ]
}

And one more question:

How do you position your splishsplash simulator?
Does this simulator belong to the category of 3D graphics or to the category of scientific programs for calculations?
That is, is splishsplash the same fluid simulator like realflow, houdini, blender elbeem, blender mantaflow, blender flip fluids?
Or all the same, the goals of this simulator differ from those listed above?

[janbender]

What do you mean with "long time"?
Since you use a fixed small time step size, the simulation will be rather slow. You also use very small values for maxError and maxErrorV, which will increase the number of required pressure solver iterations significantly. A simulation with an implicit viscosity is always slower since a linear system for the viscosity has to be solved. Finally, you enabled the vorticity solver which has not much effect since a highly viscous fluid is typically not turbulent.

[PavelBlend]

And how long did it take to calculate this simulation? And on which processor?
https://youtu.be/elZieJNBYqk?t=160

[janbender]

I think the simulation took a few hours on two Xeon processors.

However, you cannot compare FLIP particles with SPH particles. SPH solves the Navier-Stokes equations on the particle discretization while FLIP solves it on a fixed grid and only uses the particles to transport the quantities. It is easy to simulate millions of FLIP particles even in real-time. Take a grid with just one cell and put all inside. However, this gives you no more details. So you should compare the number of cells with the number of SPH particles.

That said SPlisHSPlasH is typically not that slow. Did you build the library in "Release" mode? Does it run in parallel. Can you give me some simulation times?

[mskr]

Release Mode makes a huge difference here.

(with the second option you can still debug, set breakpoints etc.)

For example, simulating until t=0.01 in DamBreakModel.json took 12.573 s in Debug Mode and 0.584 s in Release With Debug Info Mode, so you get 21.5x speedup.