SpiNN_LB is a Master thesis project. SpiNN_LB is a prototype of a basic D2Q9 lattice Boltzmann method implemented on the world-leading neural-inspired hardware archtechture, the SpiNNaker platform.
Student name: Yuan Feng
Student number: s1909558
EPCC supervisor name(s): * Dr. Kevin Stratford External supervisor(s): * Dr. Alan Stokes (University of Manchester)
The SpiNNaker is a dependency-hungry program. The whole development is under the following version of SpiNNaker software and do not guarantee to run at any other version.
| software | tag | git hash |
|---|---|---|
| spinn_common | SpiNNaker 5.0.0 v5.0 | e954a29 |
| DataSpecification | ~ | dab4cf8 |
| SpiNNFrontEndCommon | 5dab2afe | |
| SpiNNMachine | dc6d29e | |
| SpiNNMan | f6d74bf | |
| PACMAN | dae8e04e | |
| SpiNNStorageHandlers | 33494c6 | |
| SpiNNUtils | 3ecf242 | |
| sPyNNaker | 5c35967f0 | |
| spalloc | acdec43 | |
| sPyNNaker8 | 27b45f5b | |
| spalloc_server | 9f16dfe | |
| SpiNNakerGraphFrontEnd | 21d1885 | |
| spinnaker_tools | SpiNNaker 5.0.0 v3.2.5 | 12e29c5 |
If you have any trouble reflog to the git version and just want a check of the results, just use the executables in the /SpiNN_LB/executable_backup
Before compile, you need a SpiNNaker development installed, if not look here: https://spinnakermanchester.github.io/ .
cd SpiNN_LB
makeAnd there would be two generated files: lattice_cell.aplx and lattice_cell.dict
Then if you have SpiNNaker board (if you do have a board you probably know how to run), run the python code in the /SpiNN_LB/pythonCode, especiallylattice_partitioned.py. However, this project is not developed with a physical board on the hand. Therefore, the lattice_partitioned.py do not guarantee to work.
Most of you may not have a board. Take look at spinn-20.cs.man.ac.uk/ you can connect to a SpiNNaker machine there in a Jupyter Notebook.
Then upload the lattice_cell.aplx and lattice_cell.dict and create a Jupyter Notebook with SpyNNaker kernel. Then run the code in /SpiNN_LB/pythonCode/LB_method.ipynb.
These code implement a 128 * 128 lattice Boltzmann method by default in Minion and Brown's paper (https://permalink.lanl.gov/object/tr?what=info:lanl-repo/lareport/LA-UR-96-4037).
If you want to run in a different scale you need to: 1. Change the kinematic viscosity (nu) and relaxation time (tau) in C according to your physical settings 2. Change the running time (in timestep) in Python according to your physical settings 3. Change the simulation scale in Python according to your physical settings 4. Probably adjust the maximum delay and time_scale_factor in Python and make it work.
In the /SpiNN_LB/pythonCode/LB_method.ipynb, we implement a contour visualizer of vorticity of the simulation. Use it.
.
├── build
├── C_implementation // A LBM implementation in C
│ └── source
│ ├── generate_pic.py // generate a contour graph for visualiztion
│ ├── Makefile
│ └──pure_c_lbm.c
├── evaluation_code
│ └── lbm_test.py
├── LICENSE
├── Makefile // Makefile compile the C
├── pythonCode
│ ├── global_settings.py
│ ├── lattice_basic_cell.py
│ ├── lattice_edge.py
│ ├── lattice_partitioned.py
│ └── LB_method.ipynb // Jupyter Notebook that run the LBM on SpiNNaker
├── README.md
└── src
└── lattice_cell.c // A basic lattice (as in lattice Boltzmann) on SpiNNaker in C