Performance characterization of video analytics workloads in heterogeneous edge infrastructures

Daniel Rivas, Francesc Guim, Jordà Polo, David Carrera
2021 Concurrency and Computation  
Powered by deep learning, video analytic applications process millions of camera feeds in real-time to extract meaningful information from their surroundings. And this number grows by the minute. To avoid saturating the backhaul network and provide lower latencies, a distributed and heterogeneous edge cloud is postulated as a key enabler for widespread video analytics. This article provides a complete characterization of end-to-end video analytics across a set of hardware platforms and
more » ... neural network architectures. Each platform is selected to fill a different gap in a distributed, shared, and heterogeneous infrastructure. Moreover, we analyze how performance scales on each of these platforms with respect to the amount of resources dedicated to video analytics. Finally, we extract the key conclusions of the characterization to build an experimental model to estimate performance and cost of end-to-end video analytics in different edge scenarios. Our experiments show that managing video analytics workloads efficiently requires awareness of both, the platforms in which these are executed, and the full end-to-end pipeline. To the best of our knowledge, this is the first work that provides a complete characterization of end-to-end video analytics in heterogeneous edge platforms. K E Y W O R D S DNN, edge cloud, end-to-end video analytics, inference, video analytics, video decoding INTRODUCTION Computer vision, thanks in part to the rapid advancements in deep learning, has boosted the use of cameras to automate the analysis and understanding of our surroundings. Video analytics is becoming increasingly popular, but its widespread adoption presents two major challenges: 1) the amount of data it generates, and 2) its computational complexity. First, the amount of data that video analytic generates. An IP camera streaming a 1080p video generates around 0.5 MiB in 1 s; the same camera will generate 1.4 TiB of data over the span of 1 month. With the current state of the internet, where 70% of the network traffic corresponds to video, 1 adding millions of cameras upstreaming content to data centers threatens to saturate backhaul networks. Therefore, the only foreseeable solution is for data to be processed close to where it is generated by means of deploying compute resources at the edge of the network. In response to this problem, the edge cloud has already been appointed as one of the main accelerators for video analytics. 2 The edge cloud promises to alleviate pressure (and costs) of the infrastructure while, at the same time, offer improved latency and overall QoS. 3 The edge cloud is defined as a decentralized and heterogeneous cloud where compute and storage is placed toward the edge of the network. Moreover, the further we move resources toward the edge of the network, the area of coverage becomes smaller and the requirements more specific. Therefore, the mix of hardware platforms to deploy can be tailored for each specific location. This is an open access article under the terms of the Creative Commons Attribution-NonCommercial License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited and is not used for commercial purposes.
doi:10.1002/cpe.6317 fatcat:fay3qd4jibe55fppiaasixs37q