Explore the practical design considerations for wide Locally Recoverable Codes (LRCs) in large-scale storage clusters in this 17-minute conference talk from FAST '23. Delve into the challenges and benefits of implementing wide LRCs to achieve higher storage savings while maintaining reliability. Learn about novel LRC constructions, including Uniform Cauchy LRCs, which demonstrate a 33% improvement in reliability for unavailability events in Google storage clusters. Discover how these codes balance low storage overhead, efficient reconstruction, and ease of deployment. Gain insights into the subtle factors affecting wide LRC reliability, often overlooked by both theoreticians and practitioners. Understand the importance of optimizing erasure codes at exascale and how wide LRCs represent the next step in unlocking greater storage efficiency.
Practical Design Considerations for Wide Locally Recoverable Codes - LRCs
Overview
Syllabus
Intro
Data redundancy prevents data loss
Need for wide codes
Locally Recoverable Codes (LRC)
Factors affecting wide LRC reliability
Wide LRCS susceptible to multiple failu
New LRC 1 : Uniform-Cauchy LRCS
New LRC 2 : Optimal-Cauchy LRCS
Comparing different LRC constructions
Robustness against random erasures
Average repair cost for 1, 2 failures
Practical Mean-time-to-data-loss M
Summary
Taught by
USENIX
