Improving the Simulation Performance of Colored Hybrid Petri Nets by the Graphics Processing Units

Abstract

The most substantial aim in systems biology is studying and understanding biological phenomena at the system level. Toward achieving this goal, it is imperative to construct and execute accurate models to predict the behaviour of the underlying system and get more insights about the interactions between the different model components. Coloured Petri nets are a promising tool to model such biological systems, which extend the power of Petri nets by assigned colours to places. However, the sequential implementation of coloured Petri net execution is prohibitively slow. Particularly, when complex models are considered that may contain reactions or species at different scales. Here, we are more interested in an extended class Petri nets class called coloured hybrid Petri nets$\left( {\mathcal{H}\mathcal{P}{\mathcal{N}^{\mathcal{C}}}} \right)$, which can combine different components at the same model. However, speeding up the simulation of $\mathcal{H}\mathcal{P}{\mathcal{N}^{\mathcal{C}}}$ models is of a paramount importance. One direction to release this goal is to resort to parallel processing. In this paper, we use the Graphics Processing Units (GPU) to increase the efficiency of simulating coloured hybrid models, whereby the time-extensive part (the stochastic regime) is simulated on the GPU, while the deterministic simulation is kept running on the CPU. Besides, the performance of our parallel approach is compared with the sequential one.