EAGER: Characterizing and Accelerating Real-Time IoT Applications using FPGAs
Rapid advances in technology scaling have driven the development of inexpensive sensing platforms with limited compute capabilities. Deployed at the network edge, these platforms enable the instrumentation of the physical world. Using offline analytical pipelines on cloud resources, municipalities have leveraged this data to manage and even improve environmental conditions promoting public health, safety, and quality of life. However, for emergency scenarios, such as large-scale building fires where network access is limited, relying on off-site cloud resources for real-time analysis of sensor data is impractical. This project investigates how hardware acceleration can be efficiently integrated into computing platforms at the network edge, enabling the use of real-time analytical pipelines at the edge to create smart, intelligent systems without requiring access to cloud resources.
This project addresses the problem of efficiently integrating acceleration within edge devices by characterizing and analyzing the power and performance of real-time workloads with a focus on high-impact applications used by first responders. Based on this characterization, we are building analytical performance and power models for these emergent workloads. These models provide insight on prospective hardware acceleration techniques and identify the limits of acceleration given the power constraints for our specific real-time use cases. Using these models we are integrating Field-Programmable Gate Arrays (FPGAs) into prototype edge devices, developing novel FPGA images for key analytical pipelines within our first responder workloads, and evaluating the power and performance impact within real systems. This work provides unique insight into the composition of future edge devices to enable real-time analytics.