Reliability Evaluation of Fault-Tolerant Multiprocessor Systems.
Abstract
The major issues involved in modeling modern computer systems can be broadly classified into those arising form the model construction, model reduction and solution, and in the interpretation of the model solution. Modeling languages such as fault trees, the PMS notation, and Extended Stochastic Petri Nets can be valuable in simplifying the task of model construction. The goal of the languages is to provide well defined constructs to the user and let the modeling package automatically generate the details of the underlying stochastic model. The language constructs should correspond closely to the system constructs, and yet should produce a concise representation. Specifying the relevent details of the system being modeled can require a tremendous number of states to be considered (in excess of 100,000). Techniques must be developed to reduce the model to one that is computationally tractable, and then to solve the reduced model in a computationally efficient manner. Once the solution is obtained, it must be interpreted carefully. The errors introduced by the model reduction step and in the solutions must be bounded, and sensitivity of the solution with respect to input parameters should be estimated. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- May 14, 1985
- Accession Number
- ADA160234
Entities
People
- K. S. Trivedi
Organizations
- Duke University