How To Architect For The Edge | An Introduction To Edge Computing

Not all computations should be done at the edge, but some workloads, particularly those requiring high privacy (like home security footage) or extreme low latency, are ideal for it. Data transfer incurs significant costs, and sending all data to the cloud for processing is often inefficient. Computer vision is a killer app for the edge due to the need for immediate processing and reduced bandwidth usage.

Latency exists on a spectrum. Hard-constrained latencies (sub-20ms) are critical in scenarios like hospital operating rooms or autonomous vehicles, where milliseconds can mean life or death (latency critical). Other scenarios, like e-commerce websites, are latency-sensitive, where slow loading times can lead to lost customers but aren't life-threatening. Choosing where to place processing (on the device, CDN, or cloud) involves balancing cost, performance, and bandwidth benefits.

Edge architects suggest thinking of the edge as a hierarchical tree. This includes thousands of devices at the extreme edge (sensors), hundreds at the far edge (covering areas like parking lots or playgrounds), a few devices covering cities, and ultimately, large-scale public clouds. Determining where to process data in this hierarchy depends on cost, performance, and security, allowing for microservices to run at various levels based on their latency requirements.

Security at the edge is different; if devices are physically accessible, the threat model changes, requiring different security considerations beyond just firewalls. Edge devices often have very limited memory, making traditional virtualization or containerization difficult. This often necessitates embedded solutions and simpler code architectures. The speaker provides a personal example with a water sensor, prioritizing energy management over immediate processing for certain tasks by offloading complex decisions to a central home server.

Edge computing brings data processing closer to the end-user, encompassing a wide range of devices from home sensors and car sensors to large towers covering a city. It addresses the internet's original design, which was download-centric, making uploading large amounts of data like 4K video streams inefficient and costly. Edge computing promises to deliver solutions for IoT and autonomous driving among other broad use cases.