Impact of fault prediction on checkpointing strategies

This paper deals with the impact of fault prediction techniques on checkpointing strategies. We extend the classical analysis of Young and Daly in the presence of a fault prediction system, which is characterized by its recall and its precision, and which provides either exact or window-based time predictions. We succeed in deriving the optimal value of the checkpointing period (thereby minimizing the waste of resource usage due to checkpoint overhead) in all scenarios. These results allow to analytically assess the key parameters that impact the performance of fault predictors at very large scale. In addition, the results of this analytical evaluation are nicely corroborated by a comprehensive set of simulations, thereby demonstrating the validity of the model and the accuracy of the results.

💡 Research Summary

The paper investigates how fault‑prediction mechanisms can be incorporated into checkpointing strategies to reduce the waste caused by checkpoint overhead in large‑scale high‑performance computing systems. Starting from the classical Young‑Daly model, which determines the optimal checkpoint interval (T_{\text{opt}} = \sqrt{2C\mu}) (where (C) is the checkpoint cost and (\mu) the mean time between failures), the authors extend the analysis to account for a predictor characterized by its recall (r) (the fraction of actual failures that are predicted) and precision (p) (the fraction of predictions that correspond to real failures). Two prediction modalities are considered: (1) exact time predictions, where the failure instant is known precisely, and (2) window‑based predictions, where the predictor supplies a time interval of length (w) that may contain the failure.

For exact predictions, the authors derive a waste function that combines three contributions: checkpoint overhead ((C/T)), lost work due to failures ((T/2\mu)(1-rp)), and the cost of unnecessary checkpoints triggered by false positives ((Cr(1-p))/T). By differentiating this function with respect to the checkpoint period (T) and setting the derivative to zero, they obtain a closed‑form expression for the optimal period:
\