~/talk

Efficient Memory Overcommitment for I/O Passthrough Enabled VMs via Fine-grained...

Talkat USENIX ATC '23 · 2023-07-10..07-12
links:none on file
name:Ben Luo
links:none on file
links:none on file
slides:none on file

V-Probe implements fine-grained page metadata inspection by injecting tiny helper binaries (e.g., Page_free and Page_order) so the hypervisor reads guest struct page flags and makes reclaimed struct-page GPA ranges read-only to force a page fault on guest writes, guaranteeing DMA-safety without swapping. It reclaims a 2 MB page in about 19 microseconds and reallocates in ~60 microseconds, two orders of magnitude faster than Ballooning and IOPF-based methods, runs on x86/Linux, and shows near-baseline performance when reclaiming up to 30% memory; continuous free-page fragmentation limits effectiveness.

AI-written summary · about

  1. 0:00Memory overcommitment intro
  2. 1:49I/O passthrough and DMA
  3. 3:14Overcommitment vs I/O passthrough
  4. 4:18Existing solutions limits
  5. 8:31WaveProbe design goals
  6. 9:18WaveProbe architecture
  7. 11:45WaveProbe data manager
  8. 13:54Memory reclamation routine
  9. 15:43Reclamation sync challenges
  10. 17:52Experimental setup
  11. 18:57Performance comparison
  12. 20:47Benchmark results
  13. 21:08Conclusion
  14. 22:34Limitations and future
~/search
type to search 6,249 conferences, 138,987 speakers and 185,172 talks
~/help
cmdctrl k
search every conference, speaker and series
/
focus the filter search
h
home
c
conferences
s
speakers
e
series
t
light / dark — the status bar button does the same
?
this help
esc
close this, then clear every filter — or just the search, while you are typing in it

Ignored while you are typing. Pages without a filter rail have no search to focus.