Convertible codes: Adaptive coding for large-scale data storage
summary
Introduces Locally Repairable Convertible Codes that combine local repairability with on-the-fly parameter changes and claims practical gains: prototype evaluations on Google and Backblaze traces show up to 15% storage savings while naive re‑encoding can spike cluster IO to 100% of capacity. Gives concrete constructions (MDS requirement, super-regular Hankel constructions or Vandermonde forms), proves lower bounds, shows an read only parity blocks regime when rF ≤ rI that is access‑optimal, and explains vector‑code gains for conversion bandwidth.
AI-written summary · about
chapters (13)
- 0:00Introduction and context
- 2:03Erasure coding basics
- 4:27Redundancy configuration
- 6:27Disk failure rate analysis
- 9:50Disk adaptive redundancy
- 11:55Data temperature dynamics
- 14:58The code conversion problem
- 19:41Framework for convertible codes
- 23:31Construction of convertible codes
- 30:19Conversion cost metrics
- 33:36Optimizing conversion codes
- 39:52Bandwidth versus access cost
- 43:52Conclusion and future work
related talks
11mExploiting Combined Locality for Wide-Stripe Erasure Coding in Distributed Storage
31mMulti-version Coding and Shared Memory Emulation
10mRevisiting Network Coding for Warm Blob Storage
27mList Recovery and Local Decoding of Expander Codes
17mParaRC: Embracing Sub-Packetization for Repair Parallelization in MSR-Coded Storage
44mTutorial on List Decodable Coding Schemes and Interactive Communication in Networks (Part 2 of 2)
15mLESS is More for I/O-Efficient Repairs in Erasure-Coded Storage
57mCoding for Interactive Communication