Has anyone experience on the dynamic Smagorinsky and dynamic Mix models for compressible flows. How many times slower are they compared to laminar NS (without model). I have coded them with 2nd order central scheme and Runge-Kutta time-stepping using Top Hat filter with trapezoidal integration rule and found 2.5 and 4 times slower, respectively. I use single processor Origin2000 on uniform grid 33x33x32 points. I need some benchmarks. Thanks

The percentage computational overhead that the dynamic filtering procedure adds depends on the numerical scheme you use to integrate the basic NS equations and also (perhaps more importantly) your implementation of this scheme.

If you are using an explicit time stepping procedure, should you be able to do some FLOP counting to compute the overhead due to dynamic filtering. If the dynamic procedure is slowing you code down by more than 2.5 times, you must have a very fast integration scheme.

Thanks for your comment, but I really need an actual number of how slower the dynamic Smagorinsky and Mix compared to no-model people commonly encounter in compressible flows. I ll feel really bad if my code's performance is far below the others. As I heard that for the incompressible version the slow down factor is below 1 for the dynamic Smagorinsky, but that's probably because the basic NS needs to solve pressure Poisson's equation which bring down the slow down factor. As I mentioned, I use 4 steps Runge-Kutta for the time integration of the basic NS and the filter is uniform (cheap, grid independent). Anybody did real a posteriory calculations?