[Perf] Linux/arm64: 4 Improvements on 5/12/2026 7:19:19 PM +00:00

[performanceautofiler]

Run Information

Name	Value
Architecture	arm64
OS	azurelinux 3.0
Queue	CobaltAzureLinux
Baseline	7900769ccdc0bef7a449063c8c87233a7f246166
Compare	7a71ac4aea5dbf9d60056cbab18ce84ab8cd9b28
Diff	Diff
Configs	CompilationMode:tiered, R2RType:r2r_interpreter, RunKind:micro

Improvements in System.Formats.Tar.Tests.Perf_TarFile

Benchmark	Baseline	Test	Test/Base	Test Quality	Edge Detector	Baseline IR	Compare IR	IR Ratio
ExtractToDirectory_Stream_Async - Duration of single invocation 📝 - Benchmark Source ADX - Test Multi Config Graph	341.50 μs	225.90 μs	0.66	0.60	False
ExtractToDirectory_Path - Duration of single invocation 📝 - Benchmark Source ADX - Test Multi Config Graph	136.74 μs	92.54 μs	0.68	0.58	False
ExtractToDirectory_Stream - Duration of single invocation 📝 - Benchmark Source ADX - Test Multi Config Graph	162.72 μs	110.62 μs	0.68	0.54	False
ExtractToDirectory_Path_Async - Duration of single invocation 📝 - Benchmark Source ADX - Test Multi Config Graph	352.32 μs	240.75 μs	0.68	0.50	False

Test Report

Repro

General Docs link: https://github.com/dotnet/performance/blob/main/docs/benchmarking-workflow-dotnet-runtime.md

git clone https://github.com/dotnet/performance.git
py .\performance\scripts\benchmarks_ci.py -f net8.0 --filter 'System.Formats.Tar.Tests.Perf_TarFile*'

Details

System.Formats.Tar.Tests.Perf_TarFile.ExtractToDirectory_Stream_Async

ETL Files

Histogram

JIT Disasms

System.Formats.Tar.Tests.Perf_TarFile.ExtractToDirectory_Path

ETL Files

Histogram

JIT Disasms

System.Formats.Tar.Tests.Perf_TarFile.ExtractToDirectory_Stream

ETL Files

Histogram

JIT Disasms

System.Formats.Tar.Tests.Perf_TarFile.ExtractToDirectory_Path_Async

ETL Files

Histogram

JIT Disasms

Docs

Profiling workflow for dotnet/runtime repository
Benchmarking workflow for dotnet/runtime repository

[LoopedBard3]

🔍 Automated Triage Analysis

Summary: 1. Bisection: Run bisection on commits 7900769ccdc0...7a71ac4aea5d using RunWithBuildAtHash to isolate the improving commit. Focus on ExtractToDirectory_Stream_Async (clearest step change, highest z-score). Bisection skipped — autofile run configuration is not supported by the bisection harness.

Finding Confidence: 4/5 (analysis accuracy)
Improvement Confidence: 4/5 (likelihood of true improvement)

📊 Full Analysis (click to expand)

Issue Overview

• Issue: #72976 - [Perf] Linux/arm64: 4 Improvements on 5/12/2026 7:19:19 PM +00:00
• Type: Performance Improvement
• Date: 2026-05-12 | Queue: AzureLinux.3.Cobalt.Arm64.Perf | OS/Arch: Linux/arm64
• Commits: 7900769ccdc0 → 7a71ac4aea5d (22 commits)
• Labels: untriaged, perf-improvement, os-linux, arch-arm64, runtime-coreclr, kind-micro, compilationmode-tiered, r2rtype-r2r_interpreter, runkind-micro
• Historical Data: Available (94-95 runs per test, 414 total runs examined)

Test Analysis

Test Name	Baseline (μs)	Compare (μs)	Δ%	μ	σ	n	z	Thresh%	Noise?	Assessment
ExtractToDirectory_Stream_Async	341.50	225.90	-33.85%	340786	25193	95	-4.56	14.79%	N	Step change, stable (CV=7.39%)
ExtractToDirectory_Path	136.74	92.54	-32.32%	149415	16807	95	-3.38	22.50%	N	Drift pattern (CV=11.25%)
ExtractToDirectory_Stream	162.72	110.62	-32.02%	149935	17603	94	-2.23	23.48%	N	Drift pattern (CV=11.74%)
ExtractToDirectory_Path_Async	352.32	240.75	-31.67%	344134	25608	95	-4.04	14.88%	N	Step change, stable (CV=7.44%)

Legend: μ=mean (ns), σ=stddev (ns), n=historical runs, z=z-score, Thresh%=adaptive threshold

Finding Confidence: 4/5 - Excellent historical data (94-95 runs per test), all tests exceed adaptive thresholds with significant z-scores. Async variants show clear step changes; sync variants show drift extending a pre-existing downward trend. CV > 10% on sync tests is explained by genuine historical drift, not pure noise.

Confirmed Improvements

ExtractToDirectory_Stream_Async (-33.85%)

• Delta: 341.50 μs → 225.90 μs (-33.85%)
• Historical Context: μ=340786 ns (σ=25193, n=95); z-score=-4.56; adaptive threshold=14.79%
• Assessment: Clear step change. Historically stable (CV=7.39%), baseline near historical mean. All 9 post-change values tightly clustered around 225-270k ns — a sudden and sustained drop well outside normal variance.
• Finding Confidence: 4/5 - Excellent historical data, strong statistical signal
• Improvement Confidence: 4/5 - Direct code path match, extreme z-score, step change pattern, corroborated by 3 sibling tests
• Likely Cause: 5cd3e8432113 - "Limit metadata block size when reading TAR archives (#127602)"
  • Affected Areas: System.Formats.Tar / TAR archive reading
  • Rationale: This commit directly modifies System.Formats.Tar source files (TarHeader.Read.cs and TarHeader.cs), introducing a MaxMetadataBlockSize constant (1 MB) and changing ValidateSize() from (uint)_size > (uint)Array.MaxLength to (ulong)_size > (ulong)MaxMetadataBlockSize. While the primary intent is a correctness/security improvement (limiting metadata block sizes), the change in cast types and comparison bounds in the hot validation path during TAR extraction could have secondary performance effects. All 4 affected tests are in the same Perf_TarFile class and exercise TarFile.ExtractToDirectory variants — the only commit in the range touching this library. The commit was authored on 2026-05-12, aligning precisely with the regression date.
  • Candidates Considered:
    • 5cd3e8432113 - Direct hit: modifies System.Formats.Tar header reading logic
    • 1bfc4babd047 - Fixes races in AsyncOperation/ManualResetValueTaskSource; could contribute to async variant improvements
    • 06c12228774a - ARM64 JIT fix for vector/mask loads and stores; ARM64-specific but unlikely to affect I/O-bound Tar operations
• Verification Status: Not verified. Bisection recommended to confirm attribution, especially given the drift pattern in sync variants.

ExtractToDirectory_Path (-32.32%)

• Delta: 136.74 μs → 92.54 μs (-32.32%)
• Historical Context: μ=149415 ns (σ=16807, n=95); z-score=-3.38; adaptive threshold=22.50%
• Assessment: Drift pattern. CV=11.25% reflects genuine pre-existing downward drift (recent before-mean ~127k vs early before-mean ~152k). After values (92-115k) continue and extend the downward trajectory. The large magnitude still exceeds the adaptive threshold.
• Finding Confidence: 4/5 - Adequate data; drift pattern adds nuance but z-score is significant
• Improvement Confidence: 3/5 - Pre-existing drift reduces certainty of single-commit attribution; the improvement may be a continuation of a broader trend rather than solely caused by one commit
• Likely Cause: 5cd3e8432113 - "Limit metadata block size when reading TAR archives (#127602)"
  • Rationale: Same as above — only commit touching System.Formats.Tar. However, the drift pattern suggests the improvement for sync variants may be partially attributable to earlier changes or environment factors.
• Verification Status: Not verified. Drift pattern warrants investigation to separate committed improvement from pre-existing trend.

ExtractToDirectory_Stream (-32.02%)

• Delta: 162.72 μs → 110.62 μs (-32.02%)
• Historical Context: μ=149935 ns (σ=17603, n=94); z-score=-2.23; adaptive threshold=23.48%
• Assessment: Drift pattern. CV=11.74% driven by strong downward drift in before series (recent before-mean ~120k vs early before-mean ~153k, with some sub-100k values near the tail). After values (88-110k) are consistent with continued downward trajectory.
• Finding Confidence: 4/5 - Adequate data; z-score just above 2.0 threshold
• Improvement Confidence: 3/5 - Weakest z-score among the four tests, clear pre-existing drift reduces attribution certainty
• Likely Cause: 5cd3e8432113 - "Limit metadata block size when reading TAR archives (#127602)"
  • Rationale: Same attribution as sibling tests. The drift pattern is most pronounced here, suggesting environmental or gradual optimization factors may dominate over a single commit.
• Verification Status: Not verified.

ExtractToDirectory_Path_Async (-31.67%)

• Delta: 352.32 μs → 240.75 μs (-31.67%)
• Historical Context: μ=344134 ns (σ=25608, n=95); z-score=-4.04; adaptive threshold=14.88%
• Assessment: Clear step change. Historically stable (CV=7.44%), baseline near historical mean (~344k). All 9 post-change values tightly clustered 239-291k ns — a sudden and sustained drop well outside normal variance.
• Finding Confidence: 4/5 - Excellent historical data, strong statistical signal
• Improvement Confidence: 4/5 - Step change pattern, strong z-score, direct code path, corroborated across all 4 tests
• Likely Cause: 5cd3e8432113 - "Limit metadata block size when reading TAR archives (#127602)"
  • Rationale: Same as ExtractToDirectory_Stream_Async — clear step change aligns with commit date, direct file path match.
• Verification Status: Not verified.

Open Questions

• Historical Data: Available and sufficient (94-95 runs). CV > 10% on ExtractToDirectory_Path and ExtractToDirectory_Stream is explained by genuine historical drift, not pure noise.
• Drift vs Step Change: Async variants (Stream_Async, Path_Async) show clear step changes (CV ~7.4%), while sync variants (Path, Stream) show drift patterns (CV ~11%). This mixed pattern may indicate multiple contributing factors or that the async codepath was more directly impacted by the change.
• Mechanism Unclear: The commit changes a validation size check (Array.MaxLength → MaxMetadataBlockSize of 1MB) with a cast change (uint → ulong). The ~32% improvement across all extraction methods is larger than expected from a validation-path change alone. This warrants investigation into whether the change has secondary effects on memory allocation or I/O buffering during extraction.
• Possible Co-factors: Commit 1bfc4babd047 (AsyncOperation/ManualResetValueTaskSource race fixes) could contribute to the async test improvements specifically.
• Cross-architecture: The improvement was detected on ARM64 Cobalt. Confirming whether the same improvement appears on x64 queues would help isolate whether this is architecture-specific.

Recommended Actions

1. Bisection: Run bisection on commits 7900769ccdc0...7a71ac4aea5d using RunWithBuildAtHash to isolate the improving commit. Focus on ExtractToDirectory_Stream_Async (clearest step change, highest z-score).
2. Verification: Verify improvement with RunWithBuildAtHash on commit 5cd3e8432113 vs baseline to confirm the Tar metadata limit change is the cause.
3. Cross-queue check: Confirm the improvement holds on non-ARM64 queues to assess whether this is architecture-specific or library-level.

---

Generated by PerfTriageAgent on 2026-05-14 09:51 UTC

[BrzVlad]

Also dotnet/runtime#126904