P2CACHE: Exploring Tiered Memory for In-Kernel File Systems Caching
summary
An in-kernel cache (P2CACHE) implements a read/write-distinguishable PM+DRAM hierarchy and a lightweight operation log to persist writes immediately to Intel Optane-style persistent memory while serving reads from DRAM page cache, using a double-write and atomic tail-pointer commit to get instant durability and strong crash consistency. It is ~2000 lines of code and evaluates against Ext4, Nova and SFS, reporting up to 200x speedup for RocksDB on Ext4 and ~3× metadata improvement over Nova, showing the design works.
AI-written summary · about
chapters (16)
- 0:00Introduction
- 0:40Trends in Storage Technologies
- 1:24Software Overhead
- 2:45Storage Devices
- 3:35Question
- 4:01Minimize IO Critical Path
- 5:23Lightweight Journaling
- 6:50P2CACHE
- 7:30Design
- 9:31Demonstration
- 13:43Final Design
- 14:57Metadata Operations
- 15:37Data Operations
- 16:04Read Performance
- 16:46Real Applications
- 17:33Summary
related talks
12mRethinking File Mapping for Persistent Memory
21mStreamCache: Revisiting Page Cache for File Scanning on Fast Storage Devices
48mPersistent Memory Usage within Linux Environment
43mPromise and Pitfalls of Persistent Memory
30mMemory tiering
28mNVDIMM & Bcache: Support Non-Volatile Memory for Linux Block Layer...
31mDesign and implementation of Autocaching for CXLSSD
29mTowards an Alternative Memory Architecture