Rethinking File Mapping for Persistent Memory
summary
Finds that file mapping consumes as much as 70% of the IO path on Intel Optane DC persistent memory and shows page-cache metadata caching hurts performance. Builds HashFS: a statically sized linear-probing hash table that maps logical blocks to physical blocks, makes allocation implicit, uses parallel SIMD-like prefetch/lookups to serve multi-block requests, and bypass page cache for direct PM access. Evaluated on Strata with YCSB on LevelDB and FileBench, HashFS uses a single hash operation per lookup and boosts throughput up to 45% over Strata.
AI-written summary · about
chapters (9)
- 0:00Start
- 1:05File 10 Before PM
- 2:16File 10 After PM
- 3:33Our Contributions
- 4:40Analysis Setup
- 5:21Is Page Caching Necessary?
- 6:34Analysis Overview
- 9:56Are File Mapping Optimizations Impactful?
- 10:47Conclusions . We must re-examine file mapping specifically for PM file
related talks
19mP2CACHE: Exploring Tiered Memory for In-Kernel File Systems Caching
48mPersistent Memory Usage within Linux Environment
47mHost Performance Booster (HPB): Introduction and Current Mainline Support Status
20mObject caching on Zoned Storage
12mEvery Mapping Counts in Large Amounts: Folio Accounting
20mPercpu hashtab traversal measurement study
30mMemory tiering
44mLarge Pages in Linux