Getting the MOST out of your Storage Hierarchy with Mirror-Optimized Storage Tiering
summary
Presents Mirror-Optimized Storage Tiering (MOST), which mirrors a small hot set across tiers and routes requests with a dynamic offload ratio optimizer so no costly migrations are needed; implemented as Cerberus on CacheLib and evaluated against H-MAP, Cloyd and others. Claims up to 3.2× higher throughput and 84% fewer device writes, uses 140 MB metadata for a 2 TB hierarchy instead of 13 GB, and balances reads/writes via per-chunk subpage tracking and routing rather than background data movement.
AI-written summary · about
chapters (9)
- 0:00Storage tiering challenges
- 3:09Introducing MOST
- 4:12MOST data layout design
- 5:15Read load balancing
- 6:39Handling writes in MOST
- 9:07Avoiding partial writes
- 11:10Implementation setup
- 12:16Dynamic workload comparison
- 15:22Conclusion
related talks
19mP2CACHE: Exploring Tiered Memory for In-Kernel File Systems Caching
12mRethinking File Mapping for Persistent Memory
30mMemory tiering
38mMaximizing Heterogeneous Memory Bandwidth in Multi-Socket Systems
20mObject caching on Zoned Storage
26mCXL Memory Tiering for heterogenous computing
15mNomad: Non-Exclusive Memory Tiering via Transactional Page Migration
43mLive in a world with multiple memory types