Fig. 5

Performance of the GP-GPU aided PCSF package versus its serial counterpart. Charts show running time and related speed-ups of the original PCSF R package and the modified version where the SSSP primitive of the Boost library has been replaced with the GP-GPU based approach, named cuPCSF. Tests were performed on the human direct label PPI network by applying three score thresholds. Right-side charts show performances for a single PCSF run, while charts on the right side show executions of randomized selections. The GP-PGU based PCSF reaches speed-ups up to 9 ×. The parallelized version outperforms better on increasing the network size as well as the amount of terminal vertices. Randomization procedures introduce additional non-parallelized steps performed by the methodology, thus speed-ups reach a maximum of 5 ×