Fig. 6

Speed and memory benchmarks. One lane of an Illumina NovaSeq run was analyzed. The two letter codes describe the output layout: S for split and I for interleaved, first for library and second for segment. A Pheniqs with PAMLD generated interleaved BAM output in 3:10 h. CRAM is much more complex than BAM and employs several compression algorithms that are more aggressive than gzip to produce smaller files, so writing CRAM output took slightly longer as expected. Pheniqs scales almost linearly with the number of available cores and is at least 20 times faster than deML, which took almost 80 h to complete. B Pheniqs memory consumption increases linearly with the number of output files, since each is allocated separate IO buffers, but even with 200 output files and generous buffers Pheniqs still used less than 1 GB of memory