Developing Software for Persistent Memory
summary
NVDIMMs and new persistence instructions from Intel let code access RAM that survives reboots, and experiments simulate latency: scans with a prefetcher double latency but only drop throughput ~7%, while a B+ tree sees ~37% loss. A prototype database (simulated PM) shows a full recovery around 8 seconds, moving critical indexes into PM cuts recovery by ~2 seconds and full-PM gives near-instant recovery (sub-5 ms) for many operations. Hardware claims include up to ~6 TB per two-socket server, and use requires cache flush/fence sequences or PM libraries.
AI-written summary · about
chapters (6)
- 0:00Introduction to persistent memory
- 5:53Advantages of new memory technology
- 13:01Paradigm shift in memory usage
- 19:39Prototyping databases with persistent memory
- 28:53Future of persistent memory development
- 30:24Q&A session
related talks
43mPromise and Pitfalls of Persistent Memory
17mPreview: Persistent Memory
06mEarly experience of Utilizing Persistent Memory for Database... (Work-in-Progress Report)
52mByte-Addressable Persistent Memory and DBMS
17mBreakthrough Data-Centric Computing with a New Memory Tier
16mAerospike + Intel® persistent memory
30mLarge-Scale Practice of Persistent Memory in Alibaba Cloud
20mRevisiting Secondary Indexing in LSM-based Storage Systems with Persistent Memory
