Robust error-management and impact of throughput in solid state storage — Characterization and first system reliability model

Abstract

Solid state storage devices (SSD) have undergone very significant and widespread adoption in recent years especially in high-performance enterprise and hyperscale storage environments, which result in increasingly diverse and evolving field usage models. While the fundamental physical models for reliability of underlying non-volatile memory (NAND) may be available, physical and statistical system-level reliability models that incorporate and unify SSD architectural design with figures of merit of data-storage usage require fundamental investigation and understanding. Towards developing such fundamental understanding for the first time in the authors' knowledge, the authors investigated the impact of data transfer and throughput on the reliability of robustly designed SSDs. Thereby, a key system-level reliability metric pertaining to accelerated lifetime stress on SSDs was found, which incorporates architectural design considerations toward understanding and modeling system-level reliability. As such, an analytical model for lifetime stress acceleration was derived from empirical data and characterization. By elucidating the interaction between the physics of operational stresses with design strength in the complex interacting system of firmware and hardware inherent in solid state storage, the developed physical model of operational reliability stands to benefit robust design principles, optimized accelerated lifetime stresses, and appropriately chosen field usage models for predictable reliability.