Dependability modelling of layered systems
Olivia Das · 2004
Most distributed systems with clients and servers are constructed with a layered software architecture. In such layered systems, failure of a processor or a failure of a process in one layer can cause many other processes that depend on its services to fail, unless they can detect the failure and reconfigure to use a redundant process. The complexity of such systems has stimulated the use of a separate fault management architecture for automatic detection of software and hardware failures and reconfiguration. Since failures of such systems may incur high cost or may result in loss of life, analyzing the dependability of such systems is of major importance. This thesis has introduced a new model, called the Dependable-LQN model, for evaluating the dependability of such layered systems. The model includes the management components and their failures and interactions in the analysis. An efficient model solution strategy is described, to compute the steady-state and the instantaneous measures. The strategy exploits the idea of aggregating the states of a system into multiple classes so that all the states in a class have the same set of effective components (i.e. the components which are working and are in use). This strategy avoids the generation and solution of large Markovian models that is commonly used for dependability evaluation. The scalability of the model has been demonstrated by undertaking a case study of a large-scale Air Traffic Control system.