Proposal of Peak Shift Method to Improve Data Durability for Large-scale Data Infrastructure#
Takaki Nakamura (Tohoku Univ.)
Abstract#
We propose a peak shift method for data protection that considers time-series changes in drive failure risk and the risk differences among drive models. The proposed method creates data protection groups of drives with different operation periods or different models so that the predicted peaks of failure risks are not concentrated at the same time. This reduces the probability of multiple drive failures occurring at the same time, thereby reducing the probability of data loss. We formulated a evaluation model for the probability of data loss under conditions where the failure risk is different for each drive and varies over time. Based on the formulated evaluation model, we evaluated the data loss probability of a data protection group with the conventional configuration method and a data protection group with the proposed method.