Lecture 9 — Storage & RAID

Physical storage media and the storage hierarchy, RAID reliability via redundancy (with a live MTTDL calculator), performance via parallelism, and RAID levels 0–6 with a live parity/recovery calculator.

Sections

🧱

Storage Media & Hierarchy

Volatile vs non-volatile, the six media, the storage-hierarchy pyramid, the disk subsystem, SAN and NAS.

🛟

RAID — Reliability & Performance

Redundancy and mirroring, the full MTTDL derivation with a live calculator, and parallelism via load balancing and striping.

💽

RAID Levels 0 – 6

Every level with its elaborated name, the worked parity examples for RAID 3/4/5/6, a live parity & recovery calculator, and how to choose a level.

Quick summary

MEDIA (fast+costly → slow+cheap): cache · main memory (VOLATILE) ·
   flash · magnetic disk · optical · tape (NON-VOLATILE).
HIERARCHY: primary (cache, main memory — volatile) · secondary/on-line (flash, disk) ·
   tertiary/off-line (tape, optical).
SAN = disk interface over a network · NAS = file-system interface over a network.

RAID = Redundant Arrays of Independent Disks: one logical disk with high capacity/speed
   (parallelism) and high reliability (redundancy). 100 disks @ MTTF 100,000 h ⇒ system MTTF ≈ 41 days.
MIRRORING (RAID 1): every write to both disks, read from either; data lost only if the second
   disk fails inside the repair window.
   λ = 1/MTTF ;  MTTDL = MTTF² / (2 × MTTR) ;  years = hours / (24 × 365).
   100,000 h & 10 h ⇒ 5×10⁸ h ≈ 57,000 years (assumes INDEPENDENT failures, instant repair).
STRIPING: bit-level (8 disks, 8× transfer, bad seek — obsolete) · block-level (block i → disk (i mod n)+1).

LEVELS: 0 block striping non-redundant · 1 mirrored · 2 memory-style ECC (Hamming) bit striping ·
   3 bit-interleaved parity · 4 block-interleaved parity (dedicated P disk) ·
   5 block-interleaved DISTRIBUTED parity · 6 P+Q dual parity (survives TWO failures).
PARITY: P = D1 ⊕ D2 ⊕ … ⊕ Dn   ·   Dk = P ⊕ (all surviving disks).
CHOICE: 2 & 4 never used (subsumed by 3 & 5) · 3 obsolete · 6 rare ·
   1 for high-update (log disks), 5 for low update rate + large data.