Practical Design Considerations for Wide Locally Recoverable Codes (LRCs)
summary
Designs produce Uniform Cauchy LRCs built by expanding Azure LRC local groups to cover parity chunks using Cauchy coefficients, constructible from existing Reed–Solomon stripes by hand. Simulation on a 96-of-105 stripe with 1,000,000 Monte Carlo trials shows higher recoverability for 7–9 simultaneous failures, lower average degraded‑read I/O for one or two failures, and a practical MTTDL improvement (≈0.96 relative but large in absolute years). Deployment replay against recorded Google unavailability events yields a 33% reduction in unavailable stripes and better robustness during maintenance.
AI-written summary · about
chapters (13)
- 0:00Intro
- 1:00Data redundancy prevents data loss
- 1:31Need for wide codes
- 2:51Locally Recoverable Codes (LRC)
- 5:03Factors affecting wide LRC reliability
- 7:05Wide LRCS susceptible to multiple failu
- 7:54New LRC 1 : Uniform-Cauchy LRCS
- 9:25New LRC 2 : Optimal-Cauchy LRCS
- 10:26Comparing different LRC constructions
- 11:46Robustness against random erasures
- 12:52Average repair cost for 1, 2 failures
- 13:37Practical Mean-time-to-data-loss M
- 15:31Summary
related talks
34mLocally Recoverable Code Constructions and Some Extensions
27mA Student's View of Locally Repairable Codes
01h 14mErasure Coding at Scale
26mLow-bandwidth recovery of linear functions of Reed-Solomon-encoded data
30mCluster storage systems gotta have HeART...
22mELECT: Enabling Erasure Coding Tiering for LSM-tree-based Storage
10mExplore Data Placement Algorithm for Balanced Recovery Load Distribution
26mFast Erasure Coding for Data Storage: A Comprehensive Study of the Acceleration...