Deep Learning is today ubiquitous and is increasingly moving from the cloud down to the edge of networked infrastructures, where it enables embedded applications to perform complex inference tasks close to the data sources, reducing long-distance data movement and alleviating the need for a powerful cloud infrastructure. Edge-class MPSoC devices featuring an on-chip FPGA fabric offer key advantages for Deep Learning inference tasks, especially for complex applications where multiple models may be run concurrently in the same platform. In this work, we propose an approach and a practical framework for the systematic characterization of multithreaded Deep Learning inference on edge FPGA MPSoCs. We instantiate the framework into a real-world MPSoC platform, targeting Xilinx Vitis-AI as a representative example of a commercial Deep Learning acceleration toolkit for edge environments. We design a comprehensive experimental campaign and apply it to the platform for several CNNs, each trained on three different datasets. We show that our approach can be used for both hardware- and software-level analysis of a target system. Among other findings, the analysis revealed a suboptimal behaviour of the underlying toolkit runtime, involving the utilization of the accelerator cores and the uneven software latency of the support library, influenced by the shapes of the input tensors.
Background. Vehicular crowdsensing (VCS) can be a cost-effective solution to gather data in urban environments, leveraging the onboard sensors of modern vehicles moving around the city. Many experimental studies have proven that high-mileage vehicles, such as taxis, can be effectively used for VCS. However, these studies have been mostly carried out in cities with regular, grid-based, road networks. Conversely, little work has been conducted to assess the suitability of VCS in cities with more complex urban road networks, such as historical ones. Goal. As a step towards filling this gap, the present study investigates the feasibility of using different-sized fleets of taxis to crowdsense information in the urban areas of the historical cities of Porto (Portugal) and Rome (Italy), whose road networks evolved over the centuries and feature a complex topology. Data and Methodology. This work leverages massive real-world datasets of taxi trajectories collected over three contiguous weeks in the cities of Porto and Rome to estimate the spatiotemporal coverage achievable by different-sized fleets of taxis if they were used for VCS. Indeed, using these trajectories, several simulations were conducted, considering four sizes of taxi fleets, ranging from 50 to 400 vehicles, for both cities. The achievable spatiotemporal road network coverage metrics were computed at a fine-grained scale of single road segments. Results. Results show that the achievable coverage in both historical cities exhibits very similar trends, with as few as 50 vehicles being capable of visiting a relevant part of the road network at least once in the considered time frame. As expected, increasing the number of involved vehicles improves spatial and temporal coverage. Still, time gaps between subsequent visits can be possibly inadequate for some VCS use cases. As a consequence, recruiting more vehicles and/or devising specialized routing/incentivization mechanisms might be necessary to achieve more comprehensive coverage of the urban road network.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.