Fig. 7
From: Bayesian localization of CNV candidates in WGS data within minutes
Comparison of benchmarks for running time, memory usage and cache behavior between the old and new versions of HaMMLET on the rat population WGS data set. The new approach yields a 17.8-fold speedup and 32.2-fold memory reduction. Notice that the number of minor page faults decreases by two orders of magnitude, indicating much better cache behavior due to the use of new data structures and an improved implementation. The number of major page faults is zero in both implementations. The wavelet tree benchmarks also contain one outlier with 2.4 billion page faults and 6.4 h runtime, likely due to undercompression. No such anomaly was observed for the breakpoint array