Optimisation, Distribution and Parallelisation
MPI 1/2
- Popular method when memory/processors are connected via a network
- Independent memory each with one of more CPUs attached
- Possibly 100,000s of CPUs
- Can also parallelise for shared memory
- Multiple processes are launched for the duration of the execution - e.g. one process per CPU
- Each process has independent memory
- Communication between processes uses MPI functions/subroutines