Skip to content

docs(audits): profile parallel stat batch size effectiveness (#1083)#3906

Merged
oferchen merged 1 commit into
masterfrom
audits/parallel-stat-batch-size-profile
May 7, 2026
Merged

docs(audits): profile parallel stat batch size effectiveness (#1083)#3906
oferchen merged 1 commit into
masterfrom
audits/parallel-stat-batch-size-profile

Conversation

@oferchen
Copy link
Copy Markdown
Owner

@oferchen oferchen commented May 7, 2026

Summary

  • Document the current per-operation parallel-stat thresholds (DEFAULT_STAT_THRESHOLD = 64 plus signature/metadata/deletion siblings) and where they dispatch via map_blocking.
  • Frame the question of whether 64 is the right cutover when NFS/FUSE/SMB stats cost 100x-1000x more than local tmpfs/xfs.
  • Lay out a profiling matrix (thresholds {16, 64, 256, 1024, 4096} x workloads {100, 1K, 10K, 100K} x filesystems tmpfs/xfs warm/xfs cold/NFSv4/FUSE) with concrete measurement methodology.
  • Enumerate decision branches (keep 64, promote to runtime-tunable behind Refactor message segments module for modularity #1554, or filesystem-aware via statfs) and risks (rayon warmup on short transfers, work-stealing imbalance on skewed dirs, NFS server overload, heterogeneous workspaces, hyperfine noise on tmpfs).

Closes #1083 once the follow-up profiling patch lands; this PR is the audit only.

Test plan

  • No code changes - documentation only.
  • Verify wc -l docs/audits/parallel-stat-batch-size-profile.md < 130.
  • Verify no references to upstream-private tooling in the audit body.

Capture the current per-operation parallel-stat thresholds, frame
the question of whether 64 still fits NFS/FUSE round-trip costs,
lay out a {16,64,256,1024,4096} x {100,1K,10K,100K} profiling
matrix across tmpfs/xfs/NFS/FUSE, and enumerate decision branches
(keep / runtime knob / statfs-aware) plus warmup, work-stealing,
NFS overload, and noise risks.
@github-actions github-actions Bot added the documentation Improvements or additions to documentation label May 7, 2026
@oferchen oferchen merged commit 34a6bd9 into master May 7, 2026
8 checks passed
@oferchen oferchen deleted the audits/parallel-stat-batch-size-profile branch May 8, 2026 03:47
oferchen added a commit that referenced this pull request May 18, 2026
…3906)

Capture the current per-operation parallel-stat thresholds, frame
the question of whether 64 still fits NFS/FUSE round-trip costs,
lay out a {16,64,256,1024,4096} x {100,1K,10K,100K} profiling
matrix across tmpfs/xfs/NFS/FUSE, and enumerate decision branches
(keep / runtime knob / statfs-aware) plus warmup, work-stealing,
NFS overload, and noise risks.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

documentation Improvements or additions to documentation

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant