On the Realisation of a Fault Tolerance Concept for Massively Parallel Systems