Edge Computing Explained: Latency, IoT, and the 5G Connection

The Speed of Light Is the First Problem

A round-trip between a device in New York and a cloud data center in Oregon takes roughly 70–100 milliseconds even over fiber-optic infrastructure — because light travels 200,000 km per second in fiber, not 300,000 km/s in vacuum, and the data must traverse routers, switches, and processing queues at each hop. For streaming video, 100ms latency is imperceptible. For a robotic surgery system, 100ms is catastrophic. For an autonomous vehicle brake response, 100ms is the difference between a near-miss and a collision. Edge computing solves the latency problem by moving computation physically closer to where data is generated and where decisions must be made — eliminating most of that distance-induced delay.

Latency: Cloud vs. Edge by the Numbers

Latency in distributed systems has two components: propagation delay (determined by physics and distance) and processing delay (determined by server load and software efficiency). Edge computing reduces both.

The physical constraint is immovable: light cannot travel faster. Edge computing accepts this constraint and reorganizes where computation happens rather than trying to make transmission faster.

IoT Architecture: From Sensors to Cloud

The Internet of Things generates data at the device layer — sensors measuring temperature, pressure, vibration, location, imagery, or biometrics. Transmitting all raw sensor data to central cloud servers is increasingly impractical at scale. A modern factory with 10,000 sensors generating 100 KB/s each produces 1 GB/s of raw data — too much to transmit, store, and process centrally in real time. Edge computing introduces local processing layers that filter, aggregate, and analyze data before it reaches the cloud.

• Device layer: Sensors and actuators. Minimal compute. Generate raw data.
• Edge layer: Local gateways, smart routers, or edge servers. Perform real-time filtering, anomaly detection, and local decision-making. Only transmit relevant processed data upstream.
• Fog layer: Intermediate compute between edge and cloud (Cisco-coined term). Aggregates from multiple edge nodes; may coordinate regional decisions.
• Cloud layer: Long-term storage, heavy analytics, model training, global coordination. Receives condensed, processed data from edge/fog layers.

In a smart manufacturing context, an edge server near a production line can detect a vibration anomaly in a motor, trigger a maintenance alert, and stop the conveyor — all within 2ms — without ever sending raw sensor data to the cloud.

5G and Ultra-Reliable Low Latency Communication (URLLC)

5G's three service categories are eMBB (enhanced mobile broadband — speed), mMTC (massive machine-type communications — IoT density), and URLLC (ultra-reliable low latency communications — mission-critical). URLLC targets end-to-end latency below 1 millisecond with 99.9999% (six nines) reliability — specifications designed specifically for industrial control, remote surgery, and autonomous vehicle coordination. URLLC is only achievable when computation occurs at the network edge (MEC nodes co-located at or near base stations), not at distant cloud data centers. 5G and edge computing are not parallel trends — they are co-dependent technologies designed together.

Autonomous Vehicles: The Sub-Millisecond Mandate

An autonomous vehicle traveling at 100 km/h (27.8 m/s) travels 2.78 centimeters every millisecond. Brake activation-to-effect takes approximately 150–300ms in modern vehicles due to mechanical latency. For the perception-decision-action loop — detecting an obstacle, deciding to brake, initiating the brake command — the processing component must be extremely fast to remain within a safe margin. This is why autonomous vehicle perception systems run locally, on dedicated AI chips (NVIDIA Drive, Mobileye EyeQ, Qualcomm Ride), not in the cloud. The cloud handles map updates, model retraining, and fleet telemetry — not real-time driving decisions. V2X (vehicle-to-everything) communication for cooperative driving between multiple vehicles adds a network latency component that URLLC 5G with edge MEC is designed to handle.

Fog Computing: The Distinctions That Matter

Fog computing, a term championed by Cisco and formalized by the OpenFog Consortium (now part of IEEE), describes a hierarchical distributed computing architecture that extends the cloud continuum from data center to device. The distinction from edge computing is one of scope: edge computing typically refers to compute at or near a single device or small local cluster; fog computing describes the entire distributed hierarchy including intermediate aggregation nodes. In practice, the terms are often used interchangeably in industry.

Security Challenges at the Edge

Distributing compute to edge nodes creates a dramatically expanded attack surface. Each edge node — potentially thousands of devices in a deployment — must be secured, updated, and monitored. Physical security is a real concern: edge nodes in factories, retail environments, or street cabinets are accessible in ways that cloud data centers behind locked cages and biometric security are not.

• Edge node compromise can affect local operations immediately, without central visibility — unlike cloud infrastructure where anomalies are centrally logged.
• Confidential computing (Intel TDX, AMD SEV) enables encrypted computation at edge nodes, protecting data even from physically compromised hardware.
• Zero-trust networking models — where every device and connection must be authenticated regardless of location — are becoming standard in enterprise edge deployments.