What is Edge Computing?
Edge computing is a method that processes data on or near the device that creates it, instead of in a distant cloud data center.
Definition
Shi and co-authors define “edge” as any computing and network resource on the path between data sources and cloud data centers. A smartphone, a home gateway and a micro data center can all be an edge. The rationale is that computing should happen close to the data sources.
NIST uses a narrower meaning. In NIST SP 500-325, edge computing is the layer of end devices and their users, with local computing on a sensor or similar device. NIST also states that no consensus existed on how edge, fog and related terms differ. Check which meaning a source uses.
Why it is used
Shi and co-authors name four concerns that edge computing can address:
- response time
- battery life
- bandwidth cost
- data safety and privacy
They give an aviation example. A Boeing 787 generates about 5 GB of data per second. The bandwidth between the aircraft and a satellite or base station is not large enough to send it all.
Hardware and model size
Edge devices often run on limited power. NVIDIA lists the Jetson AGX Orin module at up to 275 TOPS (sparse INT8) with power configurable between 15 W and 60 W. The Jetson Orin Nano is listed at up to 40 TOPS with 7 W to 15 W.
A trained neural network is often too large for such hardware. Quantization stores weights and activations in a lower-precision type, such as 8-bit integer (INT8). NVIDIA states that quantization reduces model size and accelerates computation.
Edge computing and airfield inspection
No standard or published study that we found ties edge computing to a specific airfield inspection task. Drone images of lights, obstacles and pavement can be processed on the drone, on a computer at the airfield, or in the cloud.
Frequently Asked Questions
- What is the difference between edge computing and fog computing?
- NIST describes edge computing as the layer of end devices and their users, with local computing on a sensor or similar device. Fog computing is a hierarchy of nodes between the edge and the cloud. Some authors treat the two terms as interchangeable, and NIST notes that no consensus on the distinction existed when it wrote SP 500-325 in 2018.
- Why process data at the edge?
- Processing at the edge can reduce response time, battery use, bandwidth cost and the exposure of private data. These are the four benefits named in the 2016 IEEE survey by Shi and co-authors. The benefit that matters most depends on the application.
- How does a large neural network run on small edge hardware?
- Quantization is one method. It stores model weights and activations in a lower-precision type such as 8-bit integer (INT8). NVIDIA states that this reduces model size and accelerates computation in TensorRT.
References
- shi-edge-2016
- Edge Computing: Vision and Challenges IEEE Internet of Things Journal, vol. 3, no. 5, October 2016 (W. Shi, J. Cao, Q. Zhang, Y. Li, L. Xu)
- nist-sp-500-325
- NIST Special Publication 500-325, Fog Computing Conceptual Model NIST
- nvidia-jetson-orin
- NVIDIA Jetson Orin NVIDIA
- nvidia-tensorrt-quantization
- NVIDIA TensorRT documentation, Working with Quantized Types NVIDIA
This is a reference entry. TarmacView provides drone inspection services and doesn't sell the equipment or materials described in this glossary.
