Okapi: Decoupling Data Striping and Redundancy Grouping in Cluster File Systems
summary
Okapi decouples stripes and redundancy groups, letting stripe width be tuned for IO patterns while group width is tuned for reliability and cost; prototype runs on a 20-node prototype cluster with HDD-focused design and uses modular arithmetic to infer groups, partial parities to limit client buffering, and caching/prefetching to bound degraded-read costs. Evaluation shows higher throughput across request sizes and reduces transcoding IO by ~50% (transcode avoids full rewrites by only writing new parities).
AI-written summary · about
chapters (9)
- 0:00Okapi design
- 2:42Striping and grouping background
- 8:14Decoupling stripes and groups
- 10:41Independent tuning of stripes and groups
- 12:25Performance gains from striping
- 13:08Transcoding problem in coupled systems
- 14:55Reducing transcoding IO overhead
- 15:36Building Okapi challenges
- 16:39Conclusion and future
related talks
24mSurviving a Disk Apocalypse with Single Overlap Declustered Parity
11mExploiting Combined Locality for Wide-Stripe Erasure Coding in Distributed Storage
25mCloud native data replication strategies
15mLESS is More for I/O-Efficient Repairs in Erasure-Coded Storage
21mNew Techniques to Improve Small I/O Workloads in Distributed File Systems
12mFacebook's Tectonic Filesystem: Efficiency from Exascale
42mApache Ozone - State of the Union
32mStretch Clusters in Ceph: Algorithms, Use Cases, and Improvements