Towards Development of Risk-based Checkpointing Scheme Via Parametric Bootstrapping

Optimal checkpoint placement is a commonly used technique to generate the optimal checkpoint time sequence minimizing the system cost with the recovery overhead from a system failure and the checkpoint overhead caused by check pointing itself. When the system failure occurs by an exponential distribution, especially, the corresponding optimal checkpoint interval is periodic and can be characterized by the unique solution of a non-linear equation. In this paper, we focus on statistical estimation problems and consider the probability distributions of estimators of system failure rate, optimal checkpoint interval and its associated minimum expected system cost, when the system failure time data is given in advance. Based on the parametric bootstrapping, we develop a novel approach to estimate the higher moments and confidence intervals on the above estimators. This enables us to develop a risk-based optimal checkpoint placement policy caused by the estimation error.