Edge processing with AI can increase costs related to power consumption while decreasing costs related to data movement.
Structural Costs > Edge Processing
Edge processing is a technical design pattern where processing is done at the source, rather than processing the captured information on a centralized server. The processing happens on a smaller device, with fewer resources. Via miniaturization of the elements of a traditional, large scale system, a small piece of hardware can collect and process data and then act independently of its parent systems.
Usually this design pattern requires a connection to a centralized data center but the connection can drop or be intermittent without impacting the device’s behaviour.
How is AI changing things?
Introducing AI to edge processing can require increases to hardware costs that, in tour, lower operational costs by localizing processing of data.
Cost reduction can be realized by:
- Reducing Network Traffic. AI deployed at the edge can be used to identify information of interest and send that interesting subset to a centralized server. For a system with thousands or millions of edge devices, this approach can significantly reduce costs associated with data movement and storage.
- Privacy. Filtering out private information that requires special handling at the edge reduces overall costs of information handling over the lifespan of the system.
Costs can increase when using edge processing because:
- Increasing compute to meet AI’s needs on edge devices increases hardware costs.
- Increasing the capacity of a traditional edge device will require more powerful, expensive and power-hungry hardware.
- Miniature hardware costs more than traditional hardware.
What can you do about it?
To ensure that your edge processing is designed and implemented efficiently for AI, while reducing cost and improving privacy, consider:
- Ensuring hardware is fit for your purpose. Certain processors are better suited to the application of different AI technologies and selecting the right one will manage costs.
- How much accuracy you need. As the AI is compressed to fit on smaller hardware, it loses accuracy and understanding the level of error you can tolerate is critical for a quality miniaturization process.
- The temperature. AI’s increased processing can increase the heat produced. Ensuring that your hardware is rated for the maximum temperature that the edge device will operate at is critical to ensuring reliable compute.
- Balancing data movement with power consumption. Limiting the data being sent from your edge device will decrease data movement costs, but it will also require more power as the AI works on the device to process raw information into the insights you want to centralize.
In Summary: Good engineering practices can decrease costs and enhance privacy when deploying AI at the edge.
Hi
Support this Work
You can purchase the book on Amazon:
You can listen to the podcasts for free through Patreon, Apple Podcasts or YouTube Music: