A shared memory implementation of pipelined Parareal
share
The paper introduces an OpenMP implementation of pipelined Parareal and compares it to a standard MPI-based implementation. Both versions yield essentially identical runtimes, but, depending on the compiler, the OpenMP variant consumes about 7 significantly smaller memory footprint. The higher implementation complexity, including manual control of locks, might make it difficult to use in legacy codes, though.
READ FULL TEXT