CAJ | 학술논문

针对容错操作系统的可靠性评测问题，提出一种向Linux内核注入“单个位”故障的方法。基于x86体系结构中的软件中断机制，首先利用可加载内核模块机制在时钟中断中设置一个探测点，然后将用户态的故障数据传输到内核空间，最后利用一种C/S结构的故障注入模型，分别向Linux内核的arch、fs、kernel和mm四个子系统注入故障（瞬时型、间歇型和永久型）6700余次。实验结果表明，与arch和fs子系统相比，kernel和mm子系统对故障的敏感度较高，平均检测故障覆盖率达到38.23'；与数据段相比，内核代码段对故障的敏感度较高，平均检测故障覆盖率达到73.49'，该方法提高了容错型操作系统可靠性评测的速度和准确度。
침대용착조작계통적가고성평측문제，제출일충향Linux내핵주입“단개위”고장적방법。기우x86체계결구중적연건중단궤제，수선이용가가재내핵모괴궤제재시종중단중설치일개탐측점，연후장용호태적고장수거전수도내핵공간，최후이용일충C/S결구적고장주입모형，분별향Linux내핵적arch、fs、kernel화mm사개자계통주입고장（순시형、간헐형화영구형）6700여차。실험결과표명，여arch화fs자계통상비，kernel화mm자계통대고장적민감도교고，평균검측고장복개솔체도38.23'；여수거단상비，내핵대마단대고장적민감도교고，평균검측고장복개솔체도73.49'，해방법제고료용착형조작계통가고성평측적속도화준학도。
Aiming at the problem of evaluating the dependability of fault-tolerant operating system, a scheme was proposed to inject single-bit faults in the Linux kernel. Based on software interrupt technology existing in x86 architecture, a probe point was firstly installed in the timer interrupt using loadable kernel module. Then the faults data were delivered from user space to kernel space. Finally about 6 700 faults including transient, intermittent and permanent type were conducted targeting four kernel subsystems, i. e. arch, fs, kernel and mm, by means of a C/S fault injecting model. Experiments indicate that compared with arch and fs subsystems, the kernel and mm are more sensitive to faults, the average detection fault coverage reaches 38. 23' . Compared with data segment, the code segment is more sensitive to faults, the average detection fault coverage reaches 73. 49' . This scheme improves the speed and accuracy when evaluating the dependability of fault-tolerant operating system.