Runtime and memory by input size

The benchmark uses nested subsets of the bundled assemblies containing 1, 10, 100, or 1,000 query sequences. Each subset was run through the complete SSUextract workflow with both database profiles.

Query sequences	Input nucleotides	Accepted loci	Curated wall time	Curated maximum RSS	IMG wall time	IMG maximum RSS
1	113,952	1	20.63 s	0.795 GiB	26.75 s	0.791 GiB
10	771,608	2	49.82 s	2.605 GiB	40.57 s	2.605 GiB
100	3,846,999	4	104.18 s	2.723 GiB	101.25 s	2.724 GiB
1,000	13,254,565	10	114.69 s	2.670 GiB	134.33 s	2.780 GiB

Values in the table are medians from three measured runs. The wall-time error bars show the minimum and maximum measured values.

Benchmark conditions

• One warm-up run preceded the three measured runs for each condition.
• Every run used a fresh Nextflow work directory.
• Search tasks used two threads, with an eight-CPU workflow ceiling.
• The same input subset was used for both database profiles.
• Every run produced at least one accepted locus and identical extracted sequences, summary tables, and BLAST output within its condition.

GNU time wrapped the full pixi run nextflow command. Its maximum RSS is the largest value reported for that command, not a sum across concurrent workflow processes. Runtime and memory will vary with sequence length, hit count, storage, and available CPUs.

The 32 trial records, including warm-ups, input hashes, nucleotide counts, and accepted-locus counts, are available in example_performance.tsv. The figure and summary table are generated by notebooks/example_performance.ipynb.