CBMM: Financial Advice for Kernel Memory Managers

on YouTube ↗19m

" summary

CBMM implements a kernel component called the CBMM estimator that consults empirical cost-benefit models and per-workload profiles to decide actions like huge page promotion. A Linux prototype shows soft page-fault latencies cut by 1–3 orders of magnitude, the most expensive soft faults reduced 2–3 orders, 10–1000 μs faults drop ~2 orders, and with fragmentation overall runtime improves ~35% versus stock Linux. Decisions use measured zeroing/compaction costs and profiled TLB-miss benefits at page granularity.

AI-written summary · about

" chapters (16)

1. 0:00Intro
2. 0:21Kernel Memory Management
3. 1:01Modern 1st-party Datacenter Workload
4. 2:01Low-Quality Information
5. 4:08Cost-Unaware Policies
6. 5:16Scattered implementations
7. 5:58Roadmap
8. 7:02CBMM: KEY IDEA
9. 7:42CBMM Design: Overview
10. 9:22Creating and Implementing Models
11. 11:29Example: Huge Page Cost-Benefit Model
12. 12:21Profile Generation and Maintenance
13. 14:29Implementation and Evaluation Implementation based on Linux
14. 17:08Performance (Normalized Workload Runtime)
15. 17:27Performance Normalized Workload Runt
16. 18:19Conclusion