Accelerated DEVS Simulation Using Collaborative Computation on Multi-Cores and GPUs for Fire-Spreading IoT Sensing Applications

• Published in: Applied sciences 2018-09, Vol.8 (9), p.1466
• Main Authors: Kim, Seongseop, Cho, Jeonghun, Park, Daejin
• Format: Article
• Language: English
• Subjects: Computation; Computer architecture; Computer simulation; Continuous time systems; Cores; Differential equations; discrete event simulation; Discrete time systems; GPU-based accelerated simulation; hybrid simulation; Internet of Things; Limiting factors; Power consumption; Simulation; simulation kernel distribution on multi-core architecture; Studies; Temperature
• ISSN: 2076-3417; 2076-3417
• DOI: 10.3390/app8091466

Discrete event system specification (DEVS) has been widely used in event-driven simulations for sensor-driven Internet of things (IoT) applications, such as monitoring the spread of fire disaster. Event-driven models for IoT sensor nodes and their communication is described in DEVS and they have to be integrated with continuous models of fire-spreading dynamics so that the hybrid system modeling and simulation approach have to be considered for both continuous behavior of fire-spreading and event-driven communications by large-scale IoT sensor devices. The hybrid-integrated modelling and simulation for fire-spreading in wide area and large-scale IoT devices result in more complex model evaluation, including simulation time synchronization, so that simulation acceleration is important by considering scalability in large-scale IoT-driven applications that sense fire-spreading. In this study, we proposed a scalable simulation acceleration of a DEVS-based hybrid system using heterogeneous architecture based on multi-cores and graphic processing units (GPUs). We evaluated the power consumption comparison of the proposed accelerated-simulation approach in terms of the composition of the event-driven IoT models and continuous fire-spreading models, which are tightly described in differential equations across a large number of cellular models. The demonstrated result shows that the full utilization of CPU-GPU integrated computing resources, on which event-driven models and continuous models are efficiently deployed and optimally distributed, could enable an advantage for high-performance simulation speedup in terms of execution time, although more power consumption is required, but the total energy consumption could be reduced due to fast simulation time.