Swmone is a fully-explicit finite difference non-linear shallow water model on a staggered grid, which is parallelised using domain decomposition and ghost-point communication via Julia's MPI wrapper. The 1D shallow water equations are
∂u/∂t = -u*∂u/∂x -g*∂η/∂x + ν∂²u/∂x² + F(x,t) (1)
∂η/∂t = -∂(uh)/∂x. (2)
The terms on the right-hand side of (1) are non-linear advection, pressure gradient, viscosity and forcing. The layer thickness is h = η + H, with H being constant. Equation (1) is solved in the "vector invariant" form
∂u/∂t = ∂/∂x(-1/2*u² - g*η + ν∂u/∂x) + F(x,t)
with a Runge-Kutta 4th order time integration and 2nd order centred gradients and interpolations in space.
The domain of N
grid points is split into chunks of equal size, which will be computed by individual processes separately. Each sub-domain is extend to the left and to the right with ghost points. Before every evaluation of the right-hand side the ghost points are updated via MPI.send/MPI.receive calls from adjacent processes. Only in the end a MPI.gather is executed to retrieve data from all sub-domains. Optional plotting with PyPlot (animation via updated figure).
There's a blog article, describing the model, the domain decomposition and the parallelization.
MPI and PyPlot