Fault-Tolerant Computer Architecture Modeling and Analysis

摘要

This document is the final report. It describes a three-year research effort to develop an engineering tool capable of supporting design and analysis of large-scale, parallel, fault-tolerant computer architectures (FTCAs). The tool, a software package called Computer-Aided Performance and Reliability Evaluation System (CAPRES), allows an FTCA designer or FTCA-design evaluator to predict, by analytical means, the performance, reliability, and performability of candidate computer architectures executing candidate workloads, rather than requiring costly simulation or prototype testing for such prediction, as is currently the case. The analytic capability of CAPRES will make its most important contribution early in the system engineering cycle, when many candidate designs may appear feasible and the best is far from clear. Designers using CAPRES will be able to examine in a matter of hours or days alternatives that today take weeks or months to assess using conventional methods. Keywords: Computer architecture; Fault-tolerance computing; Performance; Modeling; Computer programs; Parallel computing; Computer modeling; Computerized simulation. (kt)