Sensor fusion on the edge
Ted Shaowang, Xi Liang, Sanjay Krishnan · 2022
Due to latency and privacy concerns, we are witnessing the rise of edge computing, where computation is placed close to the point of data collection to facilitate low-latency decision making. However, we believe that a very important class of sensor fusion applications, in which data generated in a disaggregated way has to be combined to make a decision, are not well understood in the context of edge computing. The necessary data needs to be in "the right place at the right time", making intra-edge communication a significant bottleneck. In prior work, we proposed an edge-based model serving system, called EdgeServe, that not only manages a machine learning inference service, but also orchestrates data movement between nodes on an edge network. In this paper, we evaluate trade-offs in temporal synchronization between data sources, and present initial experiments that study how different knobs can affect the performance of sensor fusion applications.