Extending Layered Queueing Network with Hybrid Sub- models Representing Exceptions and Decision Making

Model elements involving decision-making present a challenge to performance analysis based on queueing formalisms, if the decisions are related to performance quantities, as in some kinds of exception decisions. In a queueing model the probability of the exception must be assumed in advance. In a state-based model the decision can be represented directly and its probability determined, however state-based models often suffer from state explosion issue which makes the model unsolvable.;This thesis creates the Hybrid Performance Modeling Methodology (HPMM). This methodology extends Layered Queueing Network (LQN) by combining it with Generalized Stochastic Petri Nets (GSPN), so that it can tackle systems with decision making and particularly with exceptions. This methodology is developed through systematic behaviour partitioning and modeling formalism selection, consistent sub-model construction with several approximation techniques, and an open tool development with the iteration solution. The tool has integrated the LQN solver and the GSPN solver so far. This methodology is further generalized as Aspect-Oriented Performance Modeling (AOPM) to provide more scalable performance analysis. HPMM is demonstrated through modeling a system with a range of resource allocation exception handling cases and systems with short and long run connection cases using the network communication protocol TCP.