Austere Flash Caching with Deduplication and Compression

on YouTube ↗20m

" summary

AustereCache claims to cut in-memory index overhead by 69.9–97.0% versus prior dedup+compress flash caches while keeping similar read hit and write-reduction ratios and high I/O throughput. It implements packetization (hash-to-slot bucketization), fixed-size sub-chunks for compressed data to avoid per-chunk length bookkeeping, and bucket-local replacement with approximate counting via count-min sketch. Prototype is a Linux user-space block device; measured CPU add is ~31.2 µs, fingerprint cost ~14.45 µs.

AI-written summary · about

" chapters (19)

1. 0:00Start
2. 0:26Flash Caching
3. 0:47Deduplication and Compression
4. 1:48Deduplicated and Compressed Flash Cache
5. 3:44Memory Amplification for Indexing
6. 5:32Our Contribution
7. 6:26Bucketization
8. 7:18(RAM) LBA-index and FP-index
9. 8:29(SSD) Metadata and Data Regions
10. 9:27Fixed-size Compressed Data Management
11. 11:27Bucket-based Cache Replacement
12. 12:49Sketch-based Reference Counting
13. 13:44Evaluation
14. 14:58Memory Overhead
15. 15:42Read Hit Ratios
16. 16:31Write Reduction Ratios
17. 17:28Throughput
18. 18:23CPU Overhead and Multi-threading
19. 19:52Conclusion