Distributed systems typically exhibit degradable performance (in the presence of faults), motivating the development of unified performance-reliability models and model-based evaluation methods. A relatively general class of such models can be formulated as reward models, where a specific reward rate is associated with each state, and one seeks to determine the total benefit derived from operating the system during some specified utilization period. This work is extended to allow the derived benefit to depend on a reward rate determined by the state and activity that led to that state, plus an additional payoff determined by each state transition. The model is then solved for the expected reward obtained during a bounded utilization period. The reward model itself is constructed from a stochastic activity network (SAN) representation of the system. An example illustrates the usefulness of these techniques.
|Original language||English (US)|
|Title of host publication||Unknown Host Publication Title|
|Number of pages||10|
|State||Published - Jan 1 1987|
ASJC Scopus subject areas