Real-time systems performance in the presence of failures

A unified methodology for modeling both soft and hard real-time systems is presented. Techniques that combine the effects of performance, reliability/availability, and deadline violation into a single model are used. An online transaction processing system is used as an example to illustrate the modeling techniques. Dynamic failures due to a transaction violating a hard deadline are taken into account by incorporating additional transitions in the Markov chain model of the failure-repair behavior. System performance in the various configurations is considered by using throughput and response-time distribution as reward rates. Since the Markov chains used in computing the distribution of response time are often very large and complex, a higher level interface based on a variation of stochastic Petri nets called stochastic reward nets is used.<>