This paper examines the problem of safely integrating independent software components, of different criticality levels, in a single system. We examine the risks of interference between independent components which share common hardware resources. We propose a definition of safe integration in which only a limited form of interference is tolerated, namely a bounded performance degradation. We show how the definition can be applied to systems modeled as input-output automata, and we compare our model to other notions of non interference and related concepts.

• dcca7.ps