摘要
在太空辐射环境中,计算机系统中的硬件部分如寄存器和存储设备中存储的数据会因单粒子效应而发生改变,可能导致程序正常执行但输出结果不正确,引发数据流错误.针对在程序级数据流错误的检测率不高并且开销大的问题,面向程序级代码,本文提出了一种基于深度学习的数据流错误检测方法 DEDDL,可智能识别关键变量.进一步针对分析结果,提出了数据流错误检测算法DAIA,可自动添加具有复算冗余和有检测功能的语句,实时检测程序的数据流错误.实验结果表明,本文提出的方法,在取得较高错误检测率的同时,相对于已有的数据流错误检测算法拥有较低的时空开销;并具有独立于具体编译器,便于实施,快速部署等优点.
In the space radiation environment,the data stored in the hardware part of the computer system,such as registers,and other storage devices,will change,which may lead to the normal execution of the program but the output result is not correct,resulting in data flow error. In order to solve the problem of low detection rate and high cost in program level data flow error detection,this paper proposes a data flow error detection method based on deep learning,DEDDL,which can identify key variables intelligently. Furthermore,according to the analysis results,a data flow error detection algorithm DAIA is proposed,which can automatically add statements with redundancy and detection function to detect the data flow errors of the program in real time. The experimental results show that the proposed method not only achieves higher error detection rate,but also has lower space-time cost compared with the existing data flow error detection algorithms,and has the advantages of independent of the specific compiler,easy implementation and rapid deployment.
作者
胡志诚
庄毅
晏祖佳
HU Zhi-cheng;ZHUANG Yi;YAN Zu-jia(College of Computer Science and Technology,Nanjing University of Aeronautics and Astronautics,Nanjing 211106,China)
出处
《小型微型计算机系统》
CSCD
北大核心
2021年第12期2680-2688,共9页
Journal of Chinese Computer Systems
基金
国家自然科学基金面上项目(61572253)资助
航空科学基金项目(2016ZC52030)资助。
关键词
数据流
软错误
时空开销
故障注入
错误检测
深度学习
data flow
soft error
time and space expenditure
fault injection
error detection
deep learning
