mpimd-csc / ParaReal.jl

A parareal orchestrator written in Julia.

ParaReal.jl

A parareal orchestrator written in Julia.

It has been tested on a Slurm allocation using 450 cores on 29 nodes.

Main features:

• Arbitrary problem and solution types
• Lazy data management (does not implicitly send data to calling process)
• Heterogeneous parareal values/iterates w.r.t. storage size and data type
• Parallel warm-up of solvers
• Logging solver start and stop times

Getting started

The package can be installed from Julia's REPL:

pkg> dev [e-mail]:jschulze/ParaReal.jl.git

Check out the demo and test files:

• demo/riccati.jl solves a differential Riccati equation (DRE) having a low-rank solution
• demo/counting.jl counts solver applications (demo for custom types)
• test/problems/diffeq-ode.jl solves a linear ordinary differential equation (ODE) using OrdinaryDiffEq.jl
• test/problems/non-diffeq.jl counts time discretization points (test for custom types)

Before running the demos, make sure to initialize the git submodules:

cd path/to/ParaReal.jl
git submodule update --init

License

The ParaReal package is licensed under MIT, see LICENSE. This does not cover files in demo/vendor/ which are redistributed under their respective licenses:

• DifferentialRiccatiEquations.jl: MIT
• Rail371.mat within DifferentialRiccatiEquations.jl: CC-BY-4.0