Research On Software Fault-tolerant Technology Based On Dual Redundant Processes

As semiconductor devices and technology improve continuously, the performance of themicroprocessor of the computer is increasingly high and the power consumption isincreasingly low. The cost is the reduction of transistor threshold voltage and noise margin,which makes the processor of the computer system more susceptible to the externalenvironment. And in some special fields, the external environment is very harsh. Thecomputer system produces a runtime error or break down easily because of the interference ofthe external environment, which could result in a serious damage which can be irreparable tothe task. In order to improve the anti-interference ability of the system to withstand the harshexternal environment, it is highly necessary to research on the fault-tolerant technology.It has a higher time overhead on existing fault-tolerant technology, and fault detectionand recovery mechanisms are studied separately. Some models only give a warning after afailure is detected and some models only discuss the recovery. Software-based fault detectionmodels currently need the support of the source file or modifying the public library files, withthe lack of user transparency.In this paper, a Process-Level Redundancy technology was used，inter-process datacomparison was set to kernel space which increased the transparency of the fault detection.Using Process-Level Redundancy technology and Checkpointing and Rollback-Recoverytechnology, a transparent software fault-tolerant model was designed. The model was able toachieve fault detection and recovery to the target program.The implementation of the model is: The task process replicated two identical processesto execute synchronously and took checkpoint regularly to save process state. Whenexecuting a system call function, the system call function parameters of the two processeswere compared. When the results were inconsistent, it recreated tasks with the checkpoint fileto continue the task. Otherwise, to ensure the completion of the task, one process executedsystem calls and the other simulated system calls. The fault-tolerant model captured systemcalls by building a kernel module, without modifying the source code and binary code,without adding the new hardware, and it was transparent to the user.