FAULT TOLERANT SYSTEM DESIGN AND EVALUATION ALGORITHMS

THE OBJECTIVE OF THIS RESEARCH IS TO INVESTIGATE NEW FAULT TOLERANT ALGORITHMS AND MODELING ENVIRONMENTS FOR USE IN RELIABILITY DESIGN AND EVALUATION OF MODEL MULTIPROCESSOR NETWORKS SUCH AS FLIGHT CONTROL SYSTEMS AND HIGH-RELIABILITY COMPUTER SYSTEMS FOR NUCLEAR PLANT SAFETY SYSTEMS AND OTHER MISSION-CRITICAL WEAPON PLATFORM COMPUTERS. EMPHASIS IS PLACED ON FAULT TOLERANCE RELATED TECHNIQUES WHICH ARE OFTEN USED IN COMPUTER NETWORKS APPLICABLE AT MULTIPLE LEVELS (E.G., CIRCUIT CARD, EQUIPMENT, SYSTEM). COMPUTATIONALLY EFFICIENT ALGORITHMS WILL BE INVESTIGATED AND EXAMPLE ALGORITHMS DESIGNED. THE INNOVATIVE ALGORITHM EVALUATION TECHNIQUE EMPLOYS AN AUTOMATED DYNAMIC NETWORK MODELING SYSTEM TO SYSTEMATICALLY ACCOMMODATE CONFIGURATION CHANGES THAT MAY OCCUR OVER TIME BECAUSE OF FAULT TOLERANCE. SYSTEM ELEMENTS AND TOPOLOGY FEATURES ARE CONSTRUCTED FROM STORED TEMPLATES WHICH PROVIDE CANONICAL REPRESENTATIONS OF COMPUTATION, COMMUNICATIONS AND FAULT TOLERANCE PROCESSES. RANDOM PROCESSES AND I/O FUNCTIONS ARE TREATED STATISTICALLY WHERE PRACTICAL, USING MOMENTS, CORRELATIONS, AND PROBABILITY DISTRIBUTIONS AS NEEDED, TO MAXIMIZE COMPUTATIONAL EFFICIENCY. MONTE CARLO TREATMENT OF RANDOM FUNCTIONS PROVIDES BACKUP UTILITY IN DEVELOPMENT OF THE STATISTICAL RELATIONSHIPS TO BE USED.