Research On Linux Kernel-Level Fault-Tolerant Technology Supporting Multi-process

With the lo wer operating voltage and the h igher frequencies in com puter chips,the probability of the transient fault in computer system is constantly growing.In som e special fields,the external environm ent is very ha rsh which m akes the proba bility increasing.In order to improve the reliability of computer sy stem,it is highly necessary to research on the fault-tolerant technology.The universality of hardware fau lt tolerant technology is poor and the cost is high.The software fault tolerance technology generally does not form a complete fault-tolerant recover mechanism and is lack of user transp arency.And all of the software fault tolerance technology in process level has latent failures problems.In this thesis,a sof tware fault tolerance mechanism in pro cess level based on L inux kernel was presented.The kernel m odule technology is used to realize the redundant comparing for increasing user transparency.The technolo gy uses the com piler debugging technology and incremental checkpoint technology to reduce the probability of latent failures and supports the m ulti-process tasks.The im plementation of the m odel is: analyzing the debugging information of the task process and transmitting the information to the kernel fault detection module,the task process replicat ed two identical processes to execute synchronously and compare the state of proc esses when executing a system call.The processes state is regu larly stored by increm ental checkpoint.The sub-proces s of the task processes is m onitored in the fault tolerance m echanism to support multi-p rocess.This mechanism does need to m odify program source code and the support of extra hardware,so the mechanism has a preferable transparency.Finally,a s oftware fault injection tool tha t cans injection multiple data segm ents of target process randomly is designed.On this ba sis,the effectiveness of reduce latent failures by the Linux kernel-level fault-tolerant technology supporting multi-process is verified.And the support of multiple process task,its parallel acceleration performance and overhead are measured.