Integration of queueing networks and Generalized Stochastic Petri Nets for the performance evaluation of computer systems
This thesis addresses the issue of being able to efficiently solve large, complex computer system models by means of a unified technique that employs a combination of product-form queueing networks and Generalized Stochastic Petri Nets. An algorithm is developed for the solution of homogeneous queueing networks with many processors and many classes. Because of the exponential complexity in terms of the number of classes for the general-purpose algorithms like Mean Value Analysis, medium to large-sized networks are beyond tractability. A particular class of queueing network is identified that is suitable for modeling local area networks; then a polynomial algorithm is developed for the solution of this category of networks that exploits the inherent symmetry of these networks. The author then describes the central ideas behind a new modeling technique that combines the best features of product form queueing networks and Generalized Stochastic Petri Nets. He explains how this technique is related to many other existing techniques. This technique has the following features: (1) a representational power and/or flexibility comparable to that of Generalized Stochastic Petri Nets: (2) an accuracy comparable to that found when using a complete set of global balance equations for the system model; and (3) an efficiency comparable to that of product-form queueing networks.
- Research Organization:
- Minnesota Univ., Minneapolis (USA)
- OSTI ID:
- 6170852
- Resource Relation:
- Other Information: Thesis (Ph. D.)
- Country of Publication:
- United States
- Language:
- English
Similar Records
Class of generalized stochastic petri nets for the performance evaluation of multiprocessor systems
'PSA-SPN' - A Parameter Sensitivity Analysis Method Using Stochastic Petri Nets: Application to a Production Line System