Lifetime Reliability-Aware Checkpointing Mechanism: Modelling and Analysis

Checkpointing mechanism is used to tolerate the impact of transient faults through roll-back operation to a previously saved system state. In this paper, we propose a novel checkpointing mechanism that considers fault tolerance in a duplex system in the presence of both transient and permanent faults. The main objective of our proposed mechanism is to extend the lifetime reliability of the duplex system by avoiding or even tolerating permanent faults in microprocessors. In addition, we also propose to migrate tasks from a 'near-to-die' processor to a spare processor under a condition where the current Mean-Time-To-Failure (MTTF) value is less or equal to a pre-determined threshold MTTF value. We validate our proposed mechanism and perform overhead analysis using various case studies. Later, we compare it with one of the most popular existing checkpointing mechanism, namely the roll-forward checkpointing scheme [9]. We show that unlike roll-back or roll-forward mechanisms, our proposed mechanism gives significantly higher lifetime reliability with reasonable system overheads.