The parallelization of a simulation problem at its bottom-level means that it is split up at the lowest possible level (Figure 6.1). In the case of the solution of a partial differential equation this could mean that the solver operates in parallel. This kind of parallelization requires a high grade of specialization and strongly depends on the nature of the problem. There exists software which supports the implementation of this kind of parallel algorithm [20,52,99]. Nevertheless, the actual implementation differs from case to case and represents a considerable effort.
Bottom-level parallelization produces maximum communication overhead compared to implementations taking place on a higher level and will therefore deliver the worst scalability. However, some situations require parallelization on bottom-level (huge simulation domain, etc.). Especially for simulations reaching to the limit of existing computers in terms of computation time and memory consumption, a parallelization on bottom-level is the only option available.
