Design And Implement Of Software-implemented Fault Injection System For Loongson3A Platform

In recent years, the technology of domestic multi-core processors is gettingbetter and better, particularly in-depth study of the Godson processors, so anincreasing number of cases began to use domestic multicore processors. Thissituation requires in-depth study of domestic multi-core processors in terms ofreliability, robustness and stability. As for this reason, we conduct a study on thereliability domestic multicore processors. Software-implement fault injection is thebasis for other reliability technologies, research on it is very important. In this paper,based on Loongson-3A platform, we design and implement a software-implementfault inject system, called LS3A-SWIFIS.In this paper, we conduct a study on Loongson-3A multi-core architecture,GS464IP core architectures and kylin operating system. Then, according to theclassical model of fault injection FARM, we proposed our fault model LS3A-FARMfor Loongson-3A. LS3A-FARM model defines the characteristics of faults set, activeset, collect set and measure set based on hardware features of Loongson-3A. Thenwe describe the way how to define four-tuple model in faults set, how to designworkloads in active set, how to get parameters in active set and how to mesure faultinject with measure set.On the basis of the definition of LS3A-FARM, the paper further describes theClient/Server design of software-implement fault injection system for Loongson-3Aplatform (LS3A-SWIFIS). On the server side, we take “how to design and implementLS3A-FARM model” and “how to implement fault injector” as two key researchtopics. As for faults set, we explain how to convert a four-tuple to a fault injection; asfor fault injector, we illustrate how to design it with system call PTRACE in detail.In addition, server-side also includes “how to implement fault detection model”,“how to implement logging model” and “how to communicate with clients”. On theclient side, we give a brief introduction to UI model and communication model.At last of this paper, we conduct experiments on Loongson-3A platfrom. Themain benchmarks include fault range/effectiveness evaluation, faults distribution andsensitivity evaluation, and performance evaluation. Fault range is the inject range ofLS3A-SWIFIS, and effectiveness is to prove the faults actually injected. Faultsensitivity illustrates reactions of the different positions on the fault injection, andfault distribution describes the distribution of faults generated after fault injection.Performance evaluation shows the effectivity of fault injection on running processes.