~/talk

MemPerf: Profiling Allocator-Induced Performance Slowdowns

Talkat OOPSLA 2023 · 2023-10-23..10-27
name:Jin Zhou
links:none on file
links:none on file
links:none on file
links:none on file
links:none on file
name:Bo Wu
links:none on file
name:Cong Liu
links:none on file
links:none on file
slides:none on file

MemPerf is a profiler that pinpoints allocator-induced slowdowns by modeling heap-object life cycles and distinguishing slow management operations from slow memory accesses. It builds an oracle from repeated runs to replace observed allocator-cycle costs, uses thread- and type-aware performance modeling plus PMU sampling and shadow memory to attribute cache and TLB misses to heap objects, and classifies miss types. The implemented tool reports 98% detection of >5% slowdowns with no false positives, imposes about 7.5% overhead, and finds real allocator design issues.

AI-written summary · about

  1. 0:00MemPerf introduction
  2. 2:36Allocator slowdown causes
  3. 4:03BS simulation modeling
  4. 5:28TLB and cache misses
  5. 8:20glibc vs tcmalloc case
  6. 9:49Evaluation against existing profilers
  7. 10:52MemPerf overhead summary
  8. 11:56Developer use cases
  9. 12:16Q&A: Scalability
  10. 13:39Q&A: Distinguishing allocator-induced misses
  11. 14:41Q&A: Allocator design
  12. 15:48Q&A: Application fixes
  13. 16:54Q&A: Architecture specifics
~/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.