Evaluating layered distributed software systems with fault-tolerant features

Proliferation of large and complex fault-tolerant distributed systems in recent years has stimulated the combined modelling of performance and dependability of such systems. For large systems it may be very expensive to compute valid performance estimates to be used in the combined performability measures. This work considers two different classes of fault-tolerant client–server systems, in which efficient layered queueing network models can be combined with AND–OR graph analysis. The fault-tolerance in the first class is achieved by incorporating stand-by servers and a mechanism to re-direct the service requests at the time of failure whereas fault-tolerance in the second class is achieved by incorporating redundant servers that can all be used at the same time with the service load being distributed equally among them. The model solutions in both the cases can be obtained without solving very large Markovian models.