Edge Computing Fundamentals
When and how to process data at the edge rather than the cloud. Latency, bandwidth, and reliability considerations for edge architecture.
Why Edge Computing?
Latency
A round trip to the cloud takes 100-500 milliseconds depending on location and network conditions. For many industrial applications - machine control, safety systems, autonomous equipment - that's far too slow. Edge processing delivers sub-millisecond response times.
Bandwidth
A single industrial sensor might generate megabytes per second. A factory with thousands of sensors produces more data than can economically be transmitted to the cloud. Edge computing filters, aggregates, and summarises data locally, sending only what's valuable.
Reliability
Network connectivity fails. When the cloud is unreachable, cloud-dependent systems stop working. Edge systems continue operating - critical for environments where downtime means safety risks or significant losses.
Data Sovereignty
Some data can't leave the premises - regulatory requirements, competitive sensitivity, or privacy concerns. Edge computing keeps data local while still enabling analytics and automation.
Edge Architecture Patterns
Device Edge
Processing happens on the device itself. Smart sensors, PLCs, and embedded systems with onboard compute capability. Provides the lowest latency but limited processing power.
Gateway Edge
A local gateway aggregates data from multiple devices, performs processing, and forwards results to the cloud. More powerful than device-level computing, still local to the facility.
On-Premises Edge
Local servers or edge appliances provide substantial compute capability. Can run complex analytics, machine learning models, or local applications. Essentially a small data centre at the facility.
Fog Computing
A distributed architecture spanning devices, gateways, and local servers. Workloads move dynamically based on requirements - latency-critical tasks stay local, batch processing goes to the cloud.
Edge-cloud continuum: Real architectures aren't purely edge or cloud - they're a continuum. Choose where to process each workload based on its specific requirements.
Industrial Edge Use Cases
Real-Time Machine Control
Feedback loops controlling industrial equipment require millisecond response. Sending data to the cloud and waiting for a response isn't viable. Edge computing enables local closed-loop control.
Predictive Maintenance
Machine learning models analyse vibration, temperature, and acoustic data to predict failures. Running inference at the edge means real-time detection without cloud round-trips.
Quality Inspection
Computer vision systems inspect products on production lines. Each inspection must complete before the next item arrives. Edge processing enables line-speed inspection.
Safety Monitoring
Safety-critical systems can't depend on cloud connectivity. Edge computing ensures safety monitoring continues even during network outages.
Data Filtering
Most sensor data is normal. Edge systems identify and transmit only anomalies, reducing bandwidth by 90%+ while ensuring important events are captured.
Design Considerations
Edge vs Cloud: What Goes Where
- Edge: Real-time control, safety systems, local analytics, data filtering
- Cloud: Historical analysis, model training, cross-site analytics, long-term storage
Hardware Selection
- Industrial ruggedisation: Temperature, humidity, vibration resistance for factory environments.
- Power considerations: Backup power for continuity during outages.
- Compute requirements: Size hardware to workload - GPUs for vision/ML, standard CPUs for data processing.
- Connectivity options: Ethernet, WiFi, cellular, industrial protocols.
Software Architecture
Edge software must handle disconnected operation, local storage, and eventual synchronisation. Key patterns:
- Store and forward: Buffer data locally during outages, sync when connectivity returns.
- Local-first operation: All critical functions work offline.
- Containerisation: Deploy and update edge workloads using containers.
- Remote management: Monitor and update edge devices centrally.
Security at the Edge
Edge devices are physically accessible to attackers. They need the same security controls as any computing device:
- Encrypted storage and communication
- Secure boot and firmware verification
- Strong authentication and access control
- Regular security updates
Edge Platforms
Cloud Provider Edge Solutions
AWS (Greengrass, Outposts), Azure (IoT Edge, Stack), and Google Cloud (Distributed Cloud) offer edge extensions of their cloud platforms. Benefits include familiar tooling and seamless cloud integration.
Industrial IoT Platforms
Siemens MindSphere, PTC ThingWorx, and GE Predix are purpose-built for industrial edge scenarios. They include industrial protocol support, historian integration, and manufacturing-specific analytics.
Open Source Options
KubeEdge, OpenNebula Edge Cloud, and EdgeX Foundry provide open-source frameworks for building edge platforms. More flexibility but require more expertise to deploy.
Summary
Edge computing is essential for industrial IoT applications with latency, bandwidth, or reliability requirements that cloud computing can't meet. The key is matching workloads to the right tier - device, gateway, local server, or cloud - based on their specific needs.
Start with clear requirements: What latency do you need? What happens during network outages? How much data are you generating? The answers drive architecture decisions. Most industrial IoT deployments end up with a hybrid approach - edge for real-time, cloud for analytics and storage.
Key takeaways
- Edge computing is about processing data where it's generated, not always in the cloud.
- Use edge for latency-critical operations, bandwidth savings, and offline resilience.
- The edge-cloud spectrum has multiple tiers: device edge, gateway, on-premises, fog, and cloud.
- Security at the edge requires secure boot, encrypted storage, and minimal attack surface.