Promise and Pitfalls of Persistent Memory
summary
Persistent memory claims real, measurable differences: 3D XPoint ships as SSDs today and DIMMs next year, offering 3 TB per socket capacity with media-level 256-byte ECC blocks, latency between DRAM and SSD, and much higher endurance than NAND. Software tooling exists—NVML on GitHub and pmemkv—for DAX/memory‑map and load/store access. Benchmarks show faster restart and WAL/Redis acceleration when properly flushing and adding transactional protection; pointer, allocator, and error‑handling changes are required and nontrivial.
AI-written summary · about
chapters (8)
- 0:00PROMISE AND PITFALLS OF PERSISTENT MEMORY
- 2:14I'M NOT ONE OF THOSE INTEL ENGINEERS INA BUNNY SUIT
- 18:52PROGRAMMING FOR PERSISTENT MEMORY
- 31:57ADAPTING SOFTWARE FOR PERSISTENT MEMORY
- 38:14POSITION INDEPENDENCE Virtual memory ranges change all the time Persistent memory ranges do not
- 38:39MEMORY ALLOCATION & GC Regular memory allocators aren't appropriate A leak in persistent memory remains leaked
- 39:04TRANSACTIONS & LOCKING Preventing torn updates using transactions Resetting abandoned locks
- 39:40ERROR DETECTION & HANDLING Coping with media failures or corruption
related talks
17mPreview: Persistent Memory
35mDeveloping Software for Persistent Memory
35mDeveloping Software for Persistent Memory
01h 08mC++ and Persistent Memory Technologies, Like Intel's 3D-XPoint
06mEarly experience of Utilizing Persistent Memory for Database... (Work-in-Progress Report)
17mBreakthrough Data-Centric Computing with a New Memory Tier
52mByte-Addressable Persistent Memory and DBMS
16mAerospike + Intel® persistent memory
