面向系统动态可靠性的自适应目标代码生成方法

Adaptive code generation method for system dynamic reliability

嵌入式系统工作在体积受限的封闭环境中,运算部件、存储单元等相关元器件体积小,集成度高,不同工作环境、不同使用频度对电子元器件的可靠性将产生重要影响。针对嵌入式系统工作过程中的动态可靠性,提出了面向系统动态可靠性的自适应目标代码生成方法。该方法借助于决策树学习算法,构建了系统可靠性评估模型;并以此为指导,设计了多路径目标代码生成方法,使得系统能够根据实际工作状态信息,自适应地执行最佳的路径,以避免系统资源使用的不均衡,提高各运算部件的可靠性。实验表明,该方法将程序对单个处理器最高使用率由80%以上降到了30%以内,将内存单元最大最小访问比例由157.3降到了15.4,有效均衡了各处理器核和内存单元的使用。

The embedded system is in the closed environment with limited volume, the computing unit, storage unit and other related components have small size and high degree of integration, so the different frequency of usage and different working en- vironment will significantly affect the reliability for the embedded system. Aiming at the dynamic reliability of the embedded system, this paper proposed a method of adaptive target code generation for the system dynamic reliability. With the help of decision tree learning algorithm, it constructed the system reliability evaluation model. Based on the reliability evaluation mo- del, it designed code generation by compiler generated codes with muhi-path, which could make the best execution path with balanced resources usages according to the actual working state information. Experiments show that this method makes the highest utilization rate of one core from higher than 80% dropped to less than 30%, and the rate of maximum and minimum access memory unit is reduced from 157.3 to 15.4, which effectively balances the use of processor cores and memory units.