改进的多版本软件扩展表决系统

Improved extended voting system for multi-version software

本文提出了一种改进的扩展表决系统,该系统是利用自动分析表决系统可靠性等参数的工具来构建的。在软件容错中采用N-模块冗余(N-Modular Redundancy-NMR)及N-版本编程(N-version Programming),应用表决技术,可以屏蔽整个软件系统中出错的部分。在实践中直接应用理论的表决模型会发现很难得到理论分析的结果。针对这样的问题,本文提出一种基于多数(plurality)的扩展表决系统:使用构造决策树的方法和进行DFT遍历作为自动化的计算工具;通过对基本的多数表决系统进行模型扩展,加入指导性规则对决策树进行重构,以自动计算工具的结果作为参考,从而选择更加可靠的表决结果作为输出。最终的模拟测试表明,所提出的扩展表决系统能够提高系统输出的正确率。

An extended plurality voting system which combines automatic analysis tool of voting system was proposed to improve performance of voting system. As a well-known technology, voting is used in fault tolerant applications of software systems to mask errors using N-Modular Redundancy （NMR） and N-version Programming. Although theoretical analysis of theoretical model of voting systems provides rigorous insight into understanding their expected behavior, theoretical model are often hard to use in practice. Therefore, we propose this extended voting system： using the automatic analysis tool which includes building decision tree and depth first traversal （DFT） algorithm, then to extending the basic plurality voting system with special rules which rebuild decision tree, finally, reference results of that tool analyzes each decision tree to select a more reliable voted result as output. Simulation experiments showed correctness rate of this extended plurality voting system＇s output was higher than basic plurality voting system.