Practical Error Control Coding For Data Storage
summary
HDD systems claim about 90% of data-center capacity and use turbo equalization with SOVA + LDPC running ~3–4 iterations per codeword, sweet-spot rates near 90% and 2–4 kB codewords; shingled magnetic recording forces read-modify-write but enables 0.25% parity overhead via track XOR parity and soft-track ECC correcting ~10% of a 400-sector track (~40 sectors). SSDs move from BCH to LDPC, TLC dominates (TLC ~3,000 P/E cycles, QLC often <1,000), use read-retry tables and multi-read soft decoding (1→~250 errors, 3→~500, 7→~550) and RAID-like XOR across dies; trellis-coded modulation is proposed for high-level (PLC) future gains.
AI-written summary · about
chapters (14)
- 0:00Introduction
- 1:04Error control coding overview
- 3:59Storage system model
- 6:25HDD latency and noise
- 8:56Turbo equalization and LDPC in HDDs
- 12:56Write squeeze and long-sector format
- 15:49Shingled recording and soft track ECC
- 23:28SSD error correction transition
- 24:33NAND flash evolution and errors
- 33:13LDPC codes and read retry in SSDs
- 43:45RAID-like protections and multi-die handling
- 46:14Reading multiple bits per cell
- 48:01Trellis-coded modulation for NAND
- 51:41NAND coding summary and challenges
related talks
15mLESS is More for I/O-Efficient Repairs in Erasure-Coded Storage
01h 14mErasure Coding at Scale
11mExploiting Combined Locality for Wide-Stripe Erasure Coding in Distributed Storage
10mRevisiting Network Coding for Warm Blob Storage
11mStripeless Data Placement for Erasure-Coded In-Memory Storage
18mDRBoost: Boosting Degraded Read Performance in MSR-Coded Storage Clusters
24mSurviving a Disk Apocalypse with Single Overlap Declustered Parity
28mUber HDFS Unit Storage Cost 10x Deduction