hesela.devRedundant array of independent NAND
RAIN uses redundancy across NAND locations inside an SSD to reconstruct data that cannot be read directly.
Protection scope: Internal XOR redundancy complements error correction within a page. It is distinct from RAID across complete drives; recovery inside an SSD cannot replace another accessible copy when that entire device is unavailable. [1] [2]
GC scheduling: TTFLASH uses parity to reconstruct a page temporarily blocked by garbage collection. Its rotating schedule normally permits one collecting plane per group, keeping a single missing stripe member recoverable. Heavy writes can require simultaneous collections, so the latency benefit is conditional. [3]
Read cost: In the paper's single-parity stripe of N pages, reconstructing a blocked partial read of R data pages adds N-R reads. For N=8 and R=1, seven additional pages must be read. These are page operations, not a latency multiplier. [3]
Relations
- related
- Error-correcting code, NAND copyback, Garbage collection (flash), NAND plane, RAID 5, Preemptible garbage collection
Limitations
- Internal protection has a finite failure scope; check the actual redundancy layout and surviving read path rather than treating the name as a whole-device guarantee. [1, 2]
- TTFLASH relies on intra-plane copyback that bypasses controller ECC checks. Its proposed background checking remained future work. [3]
Citations
- Micron. Working to revolutionize SSD resiliency with shift-left approach. History of resiliency in SSDs.
- Moon and Reddy. Don't Let RAID Raid the Lifetime of Your SSD Array. HotStorage 2013, sections 2.1 and 3.
- Yan et al. Tiny-Tail Flash: Near-Perfect Elimination of Garbage Collection Tail Latencies in NAND SSDs. FAST 2017, sections 4.2-4.4, 6.6 and 7.
Sources & caveats
- The Micron source describes a shipping internal mechanism; TTFLASH is a research design evaluated through simulation and emulation, not a property of every RAIN-equipped SSD.