Research On Techniques Of Dissimilar Fault Tolerant Computer

The fault tolerant computer can be divided into two categories: similar and dissimilar redundant computer. The similar redundant computer runs the same copy of software on each redundant computer. Because each one uses the same copy of software and the errors within the software is very hard to be avoided, the same error is copied into each computer. If an error were accidentally trigged under some conditions, all the software resided in computers would have the same fault at the same time.NVFT (N Version Fault Tolerance) is the first dissimilar computer system in China that directed by the author. Based on the engineering practice, this paper gives the summary of this project concerning with its principle and research. All works include: dissimilar redundancy computer architecture, hardware and software design, synchronization and asynchronization, communication design, software cross check point and voting vector design and the software voting and monitoring algorithm design. Also a software reliability analysis using our experimental data is presented. The conception and methods of fault insertion and flight test of the dissimilar computer are described.The diversity design principle is to use multiply design teams to independently design individual redundant computer according to the same specification. Combining with the advantages of N-version programming and recovery block, this paper presents a new fault tolerant structure — NR fault tolerant computer architecture. The analysis and implementation are also given in this paper.There are several key points to develop a N-version programming software redundancy system:1. The correctness, completeness, well-defined requirements are needed in the system specification. The specification should specify the crosscheck points and cross check vectors in detail.2. The fault tolerance system can be based on synchronization or asynchronization styles. For restraining of common noise signal, asynchronization style is a better one. This paper gives a time-tag mechanism to monitor input signals.3. In the control law design of the project, the three design teams use different algorithms to implement the specification of the crosscheck points and voting vectors.4. The software reliability analysis is an infinitude subject. This paper uses fault tree and Markov chain based on real experiment data to analyze the reliability of NVFT system.5. For the verification of a fault tolerant system, a fault inserter is developed in our project, and a new idea of the software fault injection is presented. This is very important for the system demonstration.