Turbulence Simulation

Question 1

Implicit filtering

Answer 1

• means =do nothing, the finite volume mesh automatically acts as a filter for the simulation
• used by StarCCM+ and FLUENT filtering for LES

Question 2

Explicit filtering

Answer 2

• when one needs to implement external filter functions to the simulation
• can view the test filter in the dynamic procedure as an explicit filter which is generally 2-4 times the implicit filter
• true explicit filtering is used generally with finite different codes

Question 3

HPC efficiency

Answer 3

• typically less than 100%, depends on the following
  
  1. time needed for data exchange at the boundaries of sub-domains allocated to different processors (parallel efficiency)
  2. increase in the number of inner / outer iterations necessary to fulfil the convergence criterion, due to changes in the algorithm required to parallelize it (numerical efficiency)
  3. idle time of processors when the number of control volumes per processor is uneven (load balancing efficiency)

Question 4

Energy spectrum

Answer 4

• energy containing region (largest eddies)
• inertial subrange, dissipation and wave length only important parameters
• energy dissipation region (smallest eddies), dissipation and viscosity important parameters

Question 5

Speed up efficiency

Answer 5

• typically used for the analysis of performance of parallel algorithms and comparison of algorithms and parallel computers
• Ts/Tn
• Ts is execution time for the serial algorithm
• Tn is the execution time for the parallel algorithm using n processors

Question 6

Parallel and numerical efficiency

Answer 6

PARALLEL

• can be expressed as a function of 3 main parameters
• set-up for data transfer (latency time), data transfer rate, and computing time per floating point operation

NUMERICAL
- depends on algorithms employed as some algorithms do not suffer from the numerical degradation when used on more than one processor

Question 7

Space & time parallelization

Answer 7

SPACE PARALLELIZATION

• physical computational domain is decomposed into a certain number of sub-domains
• each processor then solves for its own sub-domain and communicates only through halo cells

TIME PARALLELIZATION

• each processor is working on a different time level
• the communication requirements are huge and not ideal
• special machines and algorithms are required for time parallelization

Question 8

Smagorinsky model

Answer 8

• Cs depends on a number of factors such as natural dissipation of the code, flow geometry, boundary conditions etc
• ad-hoc wall formulations have to be implemented, such as the Van-driest wall model
• cannot account for any amount of backscatter
• dynamic smagorinsky model tries to eliminate the need for ad-hoc wall damping functions whilst accounting for some backscatter

Question 9

Wall-adapting Local Eddy Viscosity model

Answer 9

• based on true velocity gradients

- can distinguish between pure strains and rotation rates

Question 10

Speed up efficiency

Answer 10

En = Ts/n.Tn

n - no. of processors