Skip to content

Improve CentralFreeList benchmark realism.#372

Draft
copybara-service[bot] wants to merge 1 commit into
masterfrom
test_949758263
Draft

Improve CentralFreeList benchmark realism.#372
copybara-service[bot] wants to merge 1 commit into
masterfrom
test_949758263

Conversation

@copybara-service

@copybara-service copybara-service Bot commented Jul 18, 2026 •
edited
Loading

Copy link
Copy Markdown

Improve CentralFreeList benchmark realism.

  • Isolate pageheap intensive Populate from non-pageheap operations.
  • Use fleet-derived run lengths for shuffling objects into InsertRange.
  • For multi-threaded benchmark, actually use more than 1 thread.
  • Reduce harness overhead by using real PageMap class for mocking.

@copybara-service
copybara-service Bot force-pushed the test_949758263 branch 8 times, most recently from 0c0192b to 103ac6b Compare July 20, 2026 22:10
@copybara-service copybara-service Bot changed the title Refactor central freelist benchmark to use real PageMap and simulate production run lengths using telemetry-calibrated permutations. Improve CentralFreeList benchmark realism by using a deterministic fleet-derived sizeclass distribution (aligned to June 2026 fleet data) to map statically-generated spans. The objects are continuously extracted using std::shuffle and placed into a bounded FIFO so the workload dynamically churns active objects, preventing perfect CPU caching. Jul 20, 2026
@copybara-service
copybara-service Bot force-pushed the test_949758263 branch 9 times, most recently from 7360247 to 31e4730 Compare July 20, 2026 23:03
@ckennelly @copybara-github
Improve CentralFreeList benchmark realism.
0182e97 
* Isolate pageheap intensive Populate from non-pageheap operations.
* Use fleet-derived run lengths for shuffling objects into InsertRange.
* For multi-threaded benchmark, actually use more than 1 thread.
* Reduce harness overhead by using real PageMap class for mocking.

PiperOrigin-RevId: 949758263
@copybara-service copybara-service Bot changed the title Improve CentralFreeList benchmark realism by using a deterministic fleet-derived sizeclass distribution (aligned to June 2026 fleet data) to map statically-generated spans. The objects are continuously extracted using std::shuffle and placed into a bounded FIFO so the workload dynamically churns active objects, preventing perfect CPU caching. Improve CentralFreeList benchmark realism. Jul 21, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@ckennelly ckennelly Awaiting requested review from ckennelly ckennelly will be requested when the pull request is marked ready for review ckennelly is a code owner

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

1 participant

@ckennelly