Multilevel Fault Models

Chair Project Multilevel Fault Models

  • Start date: 01.01.2011
  • End date: 31.12.2014
  • Funded by: Universität Augsburg
  • Local head of project: Prof. Dr. Theo Ungerer

 

Abstract

Fault models form a central point in the design and verification of a fault-tolerant computing system. The model specifies the type and behavior of faults from the application environment that are not allowed to endanger important system properties such as safety and/or function.

 

These faults are to be detected (and tolerated) by fault-tolerance mechanisms. Metrics like reliability R(t), MTTF, MTTR, fault coverage etc. are directly determined by the model.

 

In traditional fault models, the abstraction level is fixed, e.g. gate-level (stuck-at, flip-to, stuck-open, stuck-on, bridging, crosstalk etc.). This high abstraction is advantageous if a system should be tested quickly, e.g. by fault injection but neglects effects such as fault propagation and implicitly assumes the intended application environment. Aim of this research is to model arbitrary fault effects by fault functions and to enable a mapping of a set of functions to an application environment. Operators like the application of fault to value functions or inheritance allow a structured hierarchy and the access to a set of approved fault sets in the context of the intended application environment. This provides a more accurate evaluation of fault-tolerant computing systems in terms of all known metrics.

Search