Important Information
Parallelization Options
Cache Optimization Features
Compilation and Execution Instructions
src
make
cp Solver ../working/.
cd ../working
mpirun -n [number_of_procs] ./Solver
python plot.py
Examples
1. Vortex suppression through heat transfer
Formation of von Kármán vortices behind bluff and aerodynamic bodies in a fluid flow generate structural vibrations which, especially at high Reynolds number can lead to disastrous consequences. These counter-rotating vortices are not uncommon and can easily form around objects like buildings, powerplant equipments, aircrafts, ships and submarines. The magnitude of damages is directly proportional to the frequency of vortex release which increases with the flow speed.
Figure 1. Vorticity contours around a cylinder in channel flow, t = 50 units and Re = 1000 (Red-CCW, Blue-CW) (Block grid size - 20 x 20) (Num blocks - 16 x 8)
Figure 2. Fluid properties during heat transfer for Re = 1000
From an engineering perspective, it is important to study the effect of their behavior during design stages and find a way to mitigate them. One way is to suppress the frequency of vortex release through heat transfer from the body itself. For example, consider a two-dimensional case of a solid cylinder placed inside a channel flow at Re = 1000. Fig 1(a) shows the vorticity contours with clock-wise and counter clock-wise behavior. Now, when the surface of this cylinder starts to heat up we see that these counter-rotating vortices begin to dissipate as shown in Fig 1 (b) & (c) . The vortex release frequency reduces with increase in temperature and at higher temperatures the von Kármán vortices completely disappear as in Fig 1 (d).
This behavior can be explained by observing the temperature and density distribution inside the fluid. Fig 2 shows contoured distribution of these properties for one such case. As temperature inside the fluid increases it undergoes thermal expansion which leads to a decrease in density, this causes a reduction in local Reynolds number thus suppressing the vortex formation in the wake region.
2. Lid Driven Cavity
Figure 3. (a) Numerical solution, (Block grid size - 20 x 20) (Num blocks - 6 x 6) (b) Reference solution by Ghia .et .al
Lid driven cavity is a traditional benchmark problem where the top wall is free slip and moves at a speed U and the remaining three walls are no slip. These conditions lead to interesting vortex patterns which depend on the Reynolds number. This test case is used here to validate the numerical results with reference data.
3. Heat transfer between fluid and multiple immersed boundaries
This example is to show the capability of software in handling multiple immersed boundaries. The test case here shows temperature distribution inside the fluid due to the presence of three heated cylinders.
Figure 4. Heat transfer between air and three solid cylinders, Re = 500, t = 25 units, (Block grid size - 20 x 20) (Num blocks - 40 x 40)
This type of situation is typically found in heat-exchangers and nuclear reactors and can be further understood using simulation tools.
4. MPI shared memory vs distributed memory model
Figure 5. Speed up comparison on (a) 64 core xeon phi KNL processor (b) multi - node cluster with 16 cores per node
The new MPI-3 library enables use of shared memory parallelization for processes on the same node. Whether you are running simulation on a multi-node cluster or a single MIC (Many Integrated Cores) processor you can use shared memory intra-node and distributed memory inter-node. Experimental tests for this software show that MPI shared memory parallelization gives almost twice as much speed up on Xeon Phi KNL in comparison to traditional distributed memory model. However, on a typical cluster improvement is not that significant. Efforts are underway to address the latter.