In future computing systems, such as the Internet-of-Things (IoT), functionality is increasingly defined by the networked connectivity of spatially distributed devices. This, however, poses fundamentally new design challenges and tradeoffs. Computation and communication need to be tightly coupled and jointly explored, e.g. to determine whether a functionality should be performed locally or remotely over the network in order to achieve the best performance and energy consumption. For this, designers and application developers will first and foremost need fast yet accurate simulations tools that will allow them to rapidly explore this design space. With tens to hundreds of devices in a network, traditional solutions that combine detailed system and network simulations will likely be too slow. In this work, we investigate novel networks-of-systems (NoS) simulators to close this gap. Building on our previous work on host-compiled system simulation, we develop high-level abstractions and unified models of network/system interactions. To achieve fast simulation speed, it is crucial to carefully abstract away unnecessary low-level details and describe behavior at a coarse granularity as much as possible. For improved accuracy, we employ advanced prediction and adaptivity approaches that exploit inherent knowledge about future system behavior encoded in a model. To further improve simulation speeds, we parallelize simulations on multiple host CPUs and GPUs. Results show that such simulators can scale to large numbers of nodes while providing accurate feedback about NoS performance and power consumption.
This research is funded by the Nation Science Foundation.