Synthesis of reliability models from behavioral and performance models

Current design methodologies and tools do not support the full range of system-level design and evaluation capabilities, such as integrated performance and reliability assessment, that are essential for the achievement of fault tolerance in critical applications. Besides integrating performance and reliability evaluation tools into a CAD (Computer Aided Design) environment, it is of extreme importance to develop a methodology which allows the vastly different models required in these two domains to be coupled so that changes in one are automatically reflected in the other.; The research presented here provides a modeling methodology in which a design-engineer can specify a design using a single model which is capable of encapsulating both the performance and dependability characteristics of a system. The model used by the designer is a simulation-based model in which the behavior of the system under consideration is described using a predefined set of primitive modeling constructs. The behavior and characteristics of each primitive construct are defined by a colored Petri net (CPN) which provides an unambiguous mathematical specification of the construct. Further, each primitive construct has associated with it a VHDL (VHSIC Hardware Description Language) description which has a one-to-one correspondence with the CPN defined behavior of the construct. Thus, a model built from the primitive constructs can be simulated using a VHDL simulator, and the model can be automatically converted into a CPN representation for analytical studies.; The primary objective of the research presented here is to develop techniques for the automated analysis of the system CPN model for reliability measures. The key feature of the methodology is that the designer need deal with only one model of the system from which alternate representations of the system are automatically derived. This is in contrast to existing methodologies where the user deals with multiple models of the same system. The methodology eliminates the problem of inconsistency between the different models used to perform system-level analysis and trade-offs.