Two-dimensional TMR with partial majority selection and forwarding

TMR (triple modular redundancy) is one of the most common forms of fault tolerance approaches, which is based on majority voting. Here, after each checkpoint interval, three modules are compared. If the results of at least two modules are same, the system is considered to be fault-free. In this paper, the authors propose a 2-dimensional TMR scheme (TMR-2D) and partial majority selection and forwarding scheme (PMSF) scheme which use several small sub-checkpoints instead of one large checkpoint at the voting time. With very small amount of overhead, the proposed scheme avoids many rollbacks even though the results of all three modules are different. As a result, the rollback probability and average task execution time are significantly reduced compared to the existing: schemes. The availability is also greatly improved. The proposed scheme will be effective for general fault-tolerant systems, especially for time critical systems