Fig. 4.

Scaling of runtime using multithreading based on 1000, 2000 or 3000 variants and 500 samples using simulated genotypes and phenotypes. We see that the improvement in run time with increased number of threads is not linear, due to increased overhead. In all the different runs the improvement in runtime from 5 to 40 threads is ab out 5-fold. On the other hand, the number of variants has no effect on the speed with about 9000 models per second being calculated using 40 threads across all runs. This is because as with larger data sets the individual threads handle larger data chunks at a time, leading to less overhead