Figure 2:

Performance and scalability comparison of samtools, mosdepth, and SeQuiLa-cov. Each experiment setting was repeated several times. Bar height and error bars indicate mean and range of execution time, respectively. The best pileup-based solution is definitely slower (2 times for WGS calculations) than both event-based solutions, which clearly shows the superiority of the latter one. Mosdepth execution time scales up to 5 cores; afterwards it shows no further gain in performance. SeQuiLa-cov has nearly the same execution time results as mosdepth for both block and window calculations for a single core, but scales out desirably using all 500 CPU cores on cluster nodes and at the same time performing WGS calculations in <1 minute.