GitHub

Seminar 6: Parallelization and contact

  • Concept of parallelization

  • GPU based

    • avoided, because the whole paradigm is different

CPU based

\[\begin{bmatrix}1 \\ 2 \\ 3 \\ 4 \\ 5 \\ 6 \\ 7 \end{bmatrix}_{all}\rightarrow\begin{matrix} \begin{bmatrix} 4 \\ 6 \end{bmatrix}_{C1-global}\\ \begin{bmatrix}1 \\ 2 \\ 7 \end{bmatrix}_{C2-global}\\ \begin{bmatrix}3\\5\end{bmatrix}_{C3-global}\end{matrix}\leftarrow \rightarrow\begin{matrix} \begin{bmatrix} 1 \\ 2 \end{bmatrix}_{C1-local}\\ \begin{bmatrix}1 \\ 2 \\ 3 \end{bmatrix}_{C2-local}\\ \begin{bmatrix}1\\2\end{bmatrix}_{C3-local}\end{matrix}\]

  • restructuring from fields
Data access

Not all data is avaiable at each core.

$ julia
julia> using MPI
julia> MPI.install_mpiexecjl()

Run PeriLab with two processors:

$ mpiexecjl -n 2 julia -e 'using PeriLab; PeriLab.main("examples/DCB/DCBmodel.yaml")'

Run PeriLab with one processors:

julia -e 'using PeriLab; PeriLab.main("examples/DCB/DCBmodel.yaml")'

Multi-core search strategy

In Seminar