Figure 3

Runtimes. Average time required for 100 iterations on EC2 for different amounts of input data running on data simulated with (purple) and without (teal) sequence-specific bias. In the latter case, the timing is for iterations after the first 20, which require a constant 30 minutes to learn the bias model. The cluster size is scaled as 3 slave nodes (6 cores) for each 50 million fragments. The results show that eXpress-D running on Spark maintains constant runtime when resources are scaled linearly with the amount of the data.