Model abstractions for real-time network environments

The model abstraction problem is explored from a real-time network environment perspective, usually admitting different system models (such as queue-theoretic) at its different equilibrium states. To computationally depict system states at any level of abstraction, it is necessary to identify correct models consistent with observables. However, any such system identification need not be permanent, particularly for a dynamic system. In such situations, as the system appears to migrate from one equilibrium state to another, one should be able to quickly identify an event of context-switching from one model abstraction to another. In this paper we show how, using a variation of traditional CUSUM statistical approaches, one could identify model change events on time.