Software Wear Management for Persistent Memories

on YouTube ↗32m

" summary

Kevlar is an OS-level wear-management system for byte-addressable persistent memory that uses virtual memory remapping to perform wear leveling and to migrate frequently written pages to DRAM, avoiding hardware changes. It estimates fine-grain wear without special hardware by combining Intel Precise Event Based Sampling with a software Bloom filter to track cache contents and infer write-back rates. Implemented in Linux, Kevlar reportedly extends PM lifetime (~18.4x average) with about 1.2% performance overhead.

AI-written summary · about

" chapters (17)

1. 0:00Intro
2. 0:19Promise of persistent memory (PM)
3. 2:03Wear management in software
4. 3:45Contributions
5. 4:59Wear leveling uniformly wears out PM pages
6. 8:32Wear leveling alone is not enough
7. 9:09Wear reduction in Kevlar
8. 11:27Modeling caches to measure PM writebacks
9. 13:19Kevlar approximates caches to estimate wear • Estimates temporal locality in application's access pattern • Uses bloom filter to track dirty blocks in hardware cache
10. 15:39Kevlar migrates heavily written pages to DRAM
11. 16:31Kevlar detects changes in access pattern
12. 17:00Methodology
13. 17:48Accuracy of wear estimation
14. 18:37PM device lifetime
15. 20:09Kevlar performance overhead
16. 21:00Conclusion
17. 21:55Software Wear Management for Persistent Memories