A Reliability Benchmarking Method for Linux

The main challenges faced in conducting reliability assessment of highly reliable computer products by using traditional methods are high cost, long cycle, and low au-tomation level. Although the reliability model based on small samples can acquire reliability attributes, the time expenditure remains significant. The accelerated life testing method is still in the exploratory stage due to the complexity of failure mechanisms in highly reliable computer products. This paper presents an implementation scheme of dependability bench-marking for reliability assessment of Linux operating system. The benchmarking results obtained can be used to compare the differences in reliability of different Linux distributions. The occurrence process of Linux kernel faults is modeled as Poisson process, and the timing of fault injection is determined according to the rule of exponential distribution of fault occurrence interval. The faults composing the fault injection sequence are derived from a validated fault mode library, in which the proportion of the number of fault modes of different kernel functional modules is used to characterize the spatial distribution of faults in actual situations. The selected stable operating phase software reliability model is an exponentially distributed homogeneous Poisson process model. It can produce reference values for the main reliability attributes of Linux, such as failure rate, Mean Time to Failure (MTTF), and reliability function. We also develop a tool prototype based on the proposed implementation scheme and use it for reliability assessment of CentOS Stream 8, AnolisOS-8.4-GA, openSUSE 15.3, Ubuntu 20.04.3, Fedora 35 and openEuler-20.03-LTS-SP3 Linux distributions released during the same period.