Persistent Memory Usage within Linux Environment
summary
Byte-addressable persistent memory exposes system physical ranges via ACPI and Linux DAX direct mapping, lets processes mmap files and do load/store without page cache, and requires explicit cache flush and controller flush plus fencing to guarantee durability. The NVML pmem.io library supplies transactional object stores, allocators and pool management. On emulation Redis startup drops from tens of seconds to ~1s, update throughput scales flat across instances and shows ~15x better per-instance behavior versus journaling, but full migration can cost about 50% slower compute while cutting DRAM to near zero.
AI-written summary · about
chapters (22)
- 0:00Introduction
- 0:14Persistent Memory
- 2:13Intel Persistent Memory
- 3:29Hardware vs Software
- 4:02Programming Model
- 6:22Operating System
- 7:15Memory Map 820
- 8:08Persistent Memory Usage
- 11:50Block Storage
- 13:25Data Persistentity
- 15:21Atomicity
- 17:52Positioning
- 22:08Application
- 23:36Application Engine
- 26:05Application Flow
- 26:48InMemory Database
- 30:01DRM
- 30:56Readies
- 35:09Performance per instance
- 37:48HPC World
- 38:11Matt Library
- 43:44Questions
related talks
43mLive in a world with multiple memory types
26mLet's demystify memory
22mPreserving guest memory across kexec
30mMemory tiering
19mP2CACHE: Exploring Tiered Memory for In-Kernel File Systems Caching
28mNVDIMM & Bcache: Support Non-Volatile Memory for Linux Block Layer...
11mAdvanced Parallel Memory Virtualization
12mRethinking File Mapping for Persistent Memory