Solid-State Drive

A logical overwrite is not usually a physical overwrite. In a conventional NAND SSD, the flash translation layer maps host logical block addresses to media locations. New data is programmed elsewhere and the mapping changes; recovering that mapping is part of recovering the device. NAND programming and erasure have different granularities: pages are programmed, while groups of pages are erased together as blocks. Agrawal et al. explain this architecture in sections 2.1 and 3.1.

Reclamation has a cost. Garbage collection preserves still-valid data by relocating it before erasing a victim block. Those internal writes are additional work beyond the host request. Deallocation tells a supporting device that particular logical data is no longer needed, which can reduce unnecessary copying; it is not a promise of immediate physical erasure. KIOXIA's managed-flash brief explains the mechanism; the SNIA endurance paper connects it to write amplification.

An SSD, NVMe and M.2 answer different questions. SSD identifies the device category; NVMe identifies the host/device communication architecture; M.2 identifies a physical form factor. NVM Express explicitly notes that M.2 modules can use SATA or PCIe. Matching the connector shape is therefore not enough to establish protocol compatibility.

Power-loss protection has a scope. Kingston distinguishes stored user data, volatile writes in flight and mapping metadata, and describes hardware hold-up energy together with firmware recovery. The engineering implication is that metadata recovery alone does not establish survival of every acknowledged cached write. Check the particular model and its tested configuration; do not infer that promise from the word enterprise.

Performance is a measured condition. SNIA SSS PTS 2.0.2 requires workload and preparation context when assessing steady state. Report the read/write mix, access pattern, request size, outstanding I/O, cache setting and prior writes. A short burst and a sustained workload answer different questions. The specification explicitly excludes device reliability, availability and data integrity from its scope.

Client and enterprise describe intended environments, not mandatory feature bundles. SNIA's classification distinguishes enterprise-server and enterprise-storage profiles, including different port expectations, and allows exceptions. Qualify endurance, persistence behavior and workload performance separately. The linked Hesela analyses examine those questions in more detail.