[Wei Wu (wuw92)]

Queue-only benchmarks for #1590. This PR is not intended to merge.

Results

The three-platform CI run shows gains across small, medium, and large queues. At L=2,000 with batches of 2,000-4,000 candidates, batch_insert is 4.88-23.11x faster than per-item insert, depending on the platform.

Performance depends on the workload. At L=20 with 40 candidates, batching is 23-34% slower across all three platforms. At L=2,000 with 8 candidates, it is 60% slower on Linux x64, 6% slower on Windows, and 1.52x faster on ARM64.

Speedup is median(insert) / median(batch_insert) over 27 samples per method (3 processes, 9 samples each); above 1x favors batching. Chart panels use independent x-axes. These shared-runner measurements isolate queue updates, not end-to-end search.

Raw JSON and full result tables (ZIP downloads; retained for 30 days):

• Linux x64 - AMD EPYC 9V45
• Linux ARM64 - Neoverse-N2
• Windows x64 - AMD EPYC 9V45

Run

From this PR's checkout, run the same 23-case workload without a dataset (PowerShell):

cargo build --locked --release -p diskann-benchmark
1..3 | ForEach-Object {
  .\target\release\diskann-benchmark.exe run `
    --input-file .\diskann-benchmark\example\neighbor-queue.json `
    --output-file "run-$_.json"
}
python .github\scripts\summarize_neighbor_queue.py run-1.json run-2.json run-3.json

Edit neighbor-queue.json to change queue capacity, batch size, candidate placement, or cutoff survivors. The summary script reports batch_insert / insert, so values below 1 favor batching.

[Codecov Comments Bot (codecov-commenter)]

Codecov Report

❌ Patch coverage is 91.04991% with 52 lines in your changes missing coverage. Please review.
✅ Project coverage is 92.14%. Comparing base (dd896a3) to head (a58bd09).

Files with missing lines	Patch %	Lines
diskann-benchmark/src/neighbor_queue.rs	86.05%	52 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1600      +/-   ##
==========================================
- Coverage   92.14%   92.14%   -0.01%     
==========================================
  Files         621      622       +1     
  Lines      130986   131546     +560     
==========================================
+ Hits       120694   121207     +513     
- Misses      10292    10339      +47     

Flag	Coverage Δ
miri	92.14% <91.04%> (-0.01%)	⬇️
unittests	92.09% <91.04%> (-0.01%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
diskann-benchmark/src/main.rs	94.08% <100.00%> (+0.01%)	⬆️
diskann/src/graph/index.rs	96.47% <100.00%> (-0.01%)	⬇️
diskann/src/graph/search/inline_filter_search.rs	99.52% <100.00%> (-0.01%)	⬇️
diskann/src/graph/search/multihop_filter_search.rs	100.00% <100.00%> (ø)
diskann/src/neighbor/queue.rs	99.13% <100.00%> (+0.19%)	⬆️
diskann-benchmark/src/neighbor_queue.rs	86.05% <86.05%> (ø)

... and 3 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.