A local checkpoint mechanism for on-board computing

Recent advance in the space application has increased the requirements for more real-time and reliable on-board computation. Running in a harsh space environment, on-board computers could suffer transient faults, such as single event upset (SEU) frequently. Reloading tasks or rebooting system may recover these faults, but they have seriously effects on deadlines of tasks and increase resource and energy consumption. We proposed a feasible and efficient local checkpoint model (LCM). Based on the reliability of the memory subsystem and the availability of soft-implemented fault detection techniques, a segmented rollback recovery is used in LCM. We also implemented a local checkpoint mechanism (LCMech) on VxWorks. Experiment results show that LCMech has an advantage of the space and time overhead and a strong capability of fault recovery supported by soft-implemented error detection. Therefore, the local checkpoint could satisfy requirements for real-time and reliable on-board computation and improve efficiency of fault tolerance of computer systems in space.