Models and methods for determining storage reliability

Abstract

Current dormant storage reliabiliy prediction methods are out dated and may not represent current technology. Some customers are concerned the data supporting the storage reliability prediction method are too old and not reflective of the current technology capability. This paper provides an approach and documents the results of an ongoing case study that uses a binary logistic regression (BLR) model (both classical and Bayesian) to assess recent system failures during non-operating storage and non-operating transportation. Both non-operational and operational system failures were considered in the analysis to determine presence of wear-out mechanisms and degradation, which may cause operational failures. As described in IEEE Std 1413 [1], the usefulness of a reliability prediction is based on how the prediction is developed and how well the prediction is prepared, interpreted, and applied. Reliability predictions are affected by the accuracy and completeness of the information provided to perform the prediction and the methods used to complete the prediction. The benefit of the BLR model is that it provides consistent and repeatable results that provide increased customer confidence in products.