LESS is More for I/O-Efficient Repairs in Erasure-Coded Storage
summary
LESS layers extended sub-stripes built from Reed-Solomon coding to reduce both data read and non‑contiguous seeks during single-block repair, implemented on Hadoop HDFS and tested on a local cluster and Sherman Cloud. Configurable sub-packetization (alpha) trades off data read vs number of nodes contacted: an example (n,k,alpha = 6,4,2) reads 25% less data than standard RS, about 30% of multi-block failures fall into extended-substripe cases that also speed recovery. layers extended sub-stripes configurable sub-packetization one seek per block
AI-written summary · about
chapters (11)
- 0:00Erasure-coded storage repairs
- 1:02Drawbacks of traditional codes
- 3:32Motivation for IO-efficient repair
- 4:58Introducing LESS erasure codes
- 6:22LESS repair overhead
- 8:30LESS encoding process
- 10:19Single block repair in LESS
- 11:45Multi-block repair and RS properties
- 12:27LESS implementation and evaluation
- 13:49Performance gains results
- 14:54Conclusion and Q&A
related talks
11mExploiting Combined Locality for Wide-Stripe Erasure Coding in Distributed Storage
01h 14mErasure Coding at Scale
11mStripeless Data Placement for Erasure-Coded In-Memory Storage
59mPractical Error Control Coding For Data Storage
17mParaRC: Embracing Sub-Packetization for Repair Parallelization in MSR-Coded Storage
15mSelectiveEC: Selective Reconstruction in Erasure-coded Storage Systems
16mBoosting Full-Node Repair in Erasure-Coded Storage
17mOkapi: Decoupling Data Striping and Redundancy Grouping in Cluster File Systems