A Comprehensive Analysis of Superpage Management Mechanisms and Policies

on YouTube ↗20m

" summary

Quicksilver implements reservation-based speculative allocation on FreeBSD, combining hybrid incremental-then-full preparation (promote on a utilization threshold) and accelerated zeroing via box zeroing. Evaluation across fresh and heavily fragmented machines shows stable throughput and tail latency as fragmentation rises (frag0→frag100) and reports over 2x speedup versus Linux on PageRank under fragmentation. Quicksilver also avoids memory bloat that produces >100,000 underutilized 2MB pages in a sync-heavy policy, matching or beating Linux and other systems.

AI-written summary · about

" chapters (14)

1. 0:00Intro
2. 0:26Superpages benefit large-memory Applications' performance
3. 0:53Contributions of this paper
4. 1:24x86-64 4KB-page address translation
5. 4:03Drawbacks of Superpages (2MB)
6. 5:07Five decoupled events of superpage lifetime -- To help understand OS superpage management
7. 5:53Implementation choices
8. 9:51coupling physical allocation, preparation and mapping creation brings more drawbacks System: Linux Benefit: Immediate address translation benefits and fewer page faults
9. 12:05Reservation-based policies enables speculative physical allocation, multiple page sizes and in-place promotion
10. 12:51Reservations and delaying partial deallocation fight fragmentation
11. 13:18Bulk zeroing is consistently more efficient on modern processors Typical zeroing: 512 calls of zeroing assembly code with size of 4KB Bulk zeroing: Fewer calls of zeroing assembly code with bulk size 4KB
12. 13:59Quicksilver-guided by novel observations
13. 15:31Evaluation of Quicksilver
14. 16:04Quicksilver Beats Linux on a freshly-booted machine