BCL Demultiplexing Benchmark
This is the raw Illumina run-folder barcode demultiplexing evidence track.
Current status: DotMatch supports a first lane-1 classic per-cycle BCL milestone. CBCL/NovaSeq-style input and broader multi-lane BCL operation are still gated as future work. Broader comparisons need real run folders and zero-mismatch validation against BCL Convert or bcl2fastq where available.
The public public_10x_tiny_bcl row uses the 10x Genomics Cell Ranger tiny-bcl mkfastq demo run folder. The bundled fetch script downloads the 10x sample sheet, the public run-folder archive, and the official Chromium i7 index CSV used to normalize the legacy SI-P03-C9 index-set alias into concrete index sequences.
Rows with exit code 127 are environment records for missing competitors, not runtime comparisons. They are kept in the raw table so the report is explicit about what was unavailable on this machine.
Figures
Raw Rows
tool |
workflow |
format |
clusters |
cycles |
samples |
threads |
gzip |
seconds |
clusters/sec |
peak RSS KB |
output MB |
assigned |
undetermined |
filtered |
validation |
mode |
content hash |
exit |
logs |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
dotmatch_bcl_demux |
public_10x_tiny_bcl |
classic_bcl |
2136539 |
132 |
1 |
1 |
1 |
10.029591 |
213023.5 |
297184 |
200.06 |
2079501 |
57038 |
0 |
2f4c38035535 |
0 |
../../../benchmarks/work/bcl_demux/logs/dotmatch_bcl_demux.stderr.log |
||
bcl-convert |
public_10x_tiny_bcl |
classic_bcl_or_cbcl |
2136539 |
132 |
1 |
0.000000 |
0.0 |
0 |
127 |
||||||||||
bcl2fastq |
public_10x_tiny_bcl |
classic_bcl_or_cbcl |
2136539 |
132 |
1 |
0.000000 |
0.0 |
0 |
127 |
||||||||||
cuda-demux |
public_10x_tiny_bcl |
classic_bcl_or_cbcl |
2136539 |
132 |
1 |
0.000000 |
0.0 |
0 |
127 |
Scope
Run make bcl-tiny-public-gate to verify the narrow public 10x tiny-BCL classic per-cycle milestone. This gate checks the committed DotMatch row, output hashes, count totals, and bcl2fastq count-total validation when bcl2fastq is available; an explicit not_installed environment row keeps the boundary visible without turning it into a false comparison.
Broader raw-BCL evaluation needs real classic-BCL and CBCL run-folder rows, a successful DotMatch CBCL row, competitor rows for BCL Convert/bcl2fastq/CUDA-Demux where installable, distinct repeated timing, and dotmatch bcl-validate zero-mismatch evidence.
Use make bcl-comparison-gate to check those rows when they are available.