Taming Hot Bloat Under Virtualization with HUGESCOPE
summary
HugeScope is a hypervisor-side system that detects and fixes “hot bloat” by two-phase page tracking (coarse scan then fine-grained base-page tracking) while splitting only the nested (second-level) page table, avoiding frequent VM exits and PTE metadata updates. It implements modular policies to split/coalesce based on skew, recency and pressure, adds under 4% runtime overhead, boosts tiered-memory performance up to 61% over vTMM, and saves 41% more memory than Ingens for page sharing with similar performance.
AI-written summary · about
chapters (11)
- 0:00HUGESCOPE introduction
- 1:19Huge pages in virtualization
- 3:46Hot bloat impact on memory
- 5:09Existing hot bloat limits
- 7:38HUGESCOPE design insights
- 9:05HUGESCOPE page tracking
- 11:19Page split and coalesce policies
- 12:47HUGESCOPE with tiered memory
- 13:52Experimental setup
- 15:20Results in tiered memory
- 16:26Summary