Exploring the Design Space of Page Management for Multi-Tiered Memory Systems
summary
Linux 5.3 is modified and evaluated on a two-socket system combining DRAM and Intel DCPMM to expose page placement choices. The design implements AutoTiering-CPM (conservative promotion/migration) and AutoTiering-OPM (opportunistic demotion plus background promotion), chooses victims by per-page recent-access history, reserves three DRAM pages to hide demotion latency, and measures up to ~7x speedup on Graph500 and significant gains across SPEC and graph workloads by exploiting memory-tier performance rather than locality alone.
AI-written summary · about
chapters (14)
- 0:00Intro
- 0:29Large Memory Systems
- 1:26Multi-tiered Memory Systems
- 2:43Need for Page Placement in Multi-Tiered Memory
- 4:00Need for Page Reclamation in Multi-Tiered Memory
- 5:14Exploiting Access Tier and Locality
- 7:46Demotion of Least Accessed Page
- 8:56Estimating Least Accessed Page
- 10:10Hiding Latency of Page Demotion
- 10:54Experimental Environments
- 11:13Performance Evaluation
- 12:26Effectiveness of AutoTiering-CPM
- 12:57Effectiveness of LAP classification
- 13:51Conclusion
