Automatic instruction-level recovery by duplicated instructions and checkpointing

This paper proposes a software-based technique to achieve error detection and recovery at the instruction-level. The proposed technique is based on incorporation of instructions duplication and checkpointing. Similar to the previous study, all instructions are duplicated and appropriate “check” instructions are inserted to achieve error detection. After the error detection, checkpointing is used to regain the program correct state. Two optimization schemes: checksum and live variable analysis, are introduced to reduce the performance overhead. Experimental results show that most data errors can be recovered with a relative low performance overhead.