Improving reliability of Minix3 through crash-only software

In the traditional reliable system, a simple reboot of failed service is not the optimal solution for higher reliability. This state would be lost and all clients that were using the service would be affected, if a service which maintains client-related information is rebooted blindly. Based on Minix3, we proposed a new crash-only software model which uses lightweight persistence of state, accordance of process state, transparency of faults and occasion of crash-only to mitigate the problem of state loss during a reboot. This model also significantly confines error propagation within client-related state maintained by an OS service. According to this model, we implemented a new crash-only printer driver. The experimental results show that this crash-only software can continue the unfinished task after reboot, while experiencing little temporal performance degradation.