Performance

Placeholder

This page is a placeholder. Comprehensive performance benchmarks vs STAR — across multiple genome sizes, thread counts, and storage tiers — are still in progress and will land here when the data is in.

For now, the headline points:

What we know today

• Algorithmic parity. rustar-aligner implements the same core algorithms as STAR. There’s no fundamental reason for it to be faster or slower; the constant factors come down to memory layout, allocator behaviour, and how aggressively the Rust compiler vectorises the inner loops.
• Single-threaded runtime is comparable to STAR on the yeast benchmark. Detailed numbers pending.
• Memory footprint for genome loading and alignment is in the same ballpark as STAR. The on-disk index format is identical, so the in-memory representation is similar by construction.
• Multi-threaded scaling uses Rust’s rayon for the alignment phase. Scaling is approximately linear up to the IO limit of the input/output storage.

What’s coming

The benchmark suite under construction will report:

Genome index generation

• Build time vs genome size (yeast, Drosophila, mouse, human GRCh38)
• Peak RAM during build
• Effect of --genomeSAindexNbases and --runThreadN

Alignment

• Reads/second per core, single-end and paired-end
• Multi-thread scaling curve up to 32 threads
• Comparison vs STAR with identical parameters
• Effect of two-pass mode vs single-pass on total runtime
• Effect of chimeric detection on runtime

Memory

• Peak RSS during alignment
• Effect of --limitBAMsortRAM on coordinate-sorted BAM output

Output

• Stream-to-stdout (--outStd) vs file-to-disk
• BGZF compression level vs throughput tradeoff

Reporting your own numbers

If you’ve benchmarked rustar-aligner on a real dataset, we’d love to hear about it — please open an issue with the genome, read count, hardware, parameters, and timing. Real-world data shapes the optimisation work.