ScalaCache: Scalable User-Space Page Cache Management with Software-Hardware...
summary
ScalaCache moves page‑cache management into computational storage drives (CSDs), implements a fusion FTL that merges indirection layers, and runs lock‑free per‑core cache managers with a Q index on the host to avoid CSD overload. It trims the I/O control path, bundles missing pages into single NVMe commands, and uses a GC‑aware replacement policy to avoid GC stalls. Evaluation on an emulator shows 5.12× bandwidth vs kernel page cache and 1.70× vs the prior user‑space design; code is open source.
AI-written summary · about
chapters (11)
- 0:00ScalaCache project intro
- 1:48Traditional kernel page cache limits
- 3:18Host-centric design taxes
- 5:24Software-hardware coordination insights
- 6:28ScalaCache architecture
- 7:32Lightweight cache management in CSD
- 9:41CSD resource limitations
- 10:46Reducing communication overhead
- 11:52Multi-CSD parallel processing
- 12:36Evaluation methodology
- 15:56Conclusion and future
related talks
16mJohnny Cache: the End of DRAM Cache Conflicts (in Tiered Main Memory Systems)
50mLet's talk about the cache!
44mHow to speed up your application using JCache
20mSymbiosis: The Art of Application and Kernel Cache Cooperation
01h 07mEmerging Embdedded Memories and Storage
30mCaching for Speed: Practical Performance Optimization in OpenSearch & Lucene
49mCache à la carte: a framework for in-memory caching
33mScaling KV Caches for LLMs: How LMCache + NIXL Handle Network and Storage...