Designing an Edge Computing Architecture

Edge computing provides an alternative to processing all data in the cloud exclusively. Instead, edge computing enables data to be sorted at the ‘edge’ of a network. Benefits can include reduced latency, lower use of bandwidth, better reliability, and improved security. A formal definition is: ‘Edge computing is a computing approach where data is processed close to where it is generated instead of being sent to a distant cloud or central data centre.’

Designing an edge computing architecture combines layers of sensor technology, onsite computing power, networks, and security systems. Read this guide to learn about how to design an edge computing environment in an industrial setting.

Edge, fog, and cloud is a model for a distributed computing architecture where the ability to process data sits at different distances from the data source. You can break it down like this:

• Edge layer. This is the element of the architecture that is closest to the machine that is creating data. The edge layer can be on the machine in question or inside a sensor, a PLC, an industrial PC, or an embedded system. This type of installation is characterised by extremely low latency (measured in milliseconds), real-time control, immediate detection of anomalies, filtering of raw data, and rapid safety responses. Even if the wider network fails, the edge system should continue to compute
• Fog layer. The fog layer – as the name implies – can be thought of as sitting between the edge computer and the cloud. It provides an onsite computing resource that is more powerful than the typical computing resources at the edge. The fog system may be an onsite server or server rack, or even a mini data centre. The fog computing resource will aggregate information from multiple edge devices, run advanced analytics, provide short-term storage, manage edge computers, and reduce traffic to the cloud (thereby improving bandwidth). The fog system can be thought of as a way of aggregating, normalising, and filtering data so that only the information that is needed is sent to the cloud
• Cloud layer. What’s the difference between edge vs cloud? The cloud is a centralised storage and intelligence layer. It may be a public cloud (e.g., the types of cloud service run by AWS and Azure) or situated in a private cloud at an exclusive data centre. Or it may be a hybrid cloud, featuring elements of both public and private systems. The cloud is notable for high scalability, vast storage capacities, longer latency, and its typical dependence on the internet

For many small sites, the fog layer is unnecessary. Data is captured and stored at the edge and sent to the cloud. But the fog layer becomes invaluable when you are working with very large facilities with hundreds of edge assets, processing data from multiple production lines.

Any edge computing system is dependent on hardware that’s deployed locally. When selecting industrial edge computing devices, the first factor to consider is the workload. If you’re just forwarding data from a sensor, the computing power required is less than if you’re running machine learning analytics. Using AI to detect anomalies requires more processing power than simply storing data.

Like when you purchase any computer, you will need to think about the processing power of the CPU, the amount of memory needed, and any special processing features. For example, if you are running inference AI locally, you may need an edge AI accelerator. It’s also important to consider the environment your edge system will be operating in. You may require ruggedised hardware to cope with vibration and shock, or resources that can cope with dusty environments or extremes of temperature.

You will need to ensure your operational technology can communicate with your IT systems, whether that’s via an industrial Ethernet / IP, industrial protocol such as Profinet, Wi-Fi, or cellular network (e.g., 4G or 5G). The new hardware must be secure, and each edge gateway should be treated as a potential attack vector for hackers. This means edge devices need security in the same way that your servers do. Another factor influencing resiliency is power and redundancy. For example, should you add an uninterruptible power supply?

Although it may be tempting to go for lower price systems, industrial edge computing should always prioritise resilience. If edge systems fail, you may lose not only visibility but also control of production. Security is vital.

Data synchronisation and buffering help to ensure you don’t lose data when operating edge computing. Without buffering, gaps can appear. A common strategy for industrial edge computing buffering is ‘store and forward.’ Under this model, the edge device collects data and stores it locally on a drive or database. When a connection to the wider network is available, the data is transmitted. Once the cloud has confirmed receipt of the data, the local copy is marked as synchronised. This system ensures that there is eventually consistency between the data in the cloud and the data collected by the edge devices.

You can set time limits for buffering. The buffering could last for 24 hours, a week, or longer. Alternatively, you could choose to buffer based on the type of data. For example, if there is an alert or an anomaly, you may want to ensure that this data is buffered. Some systems require storage of all raw data locally. This may require a large local drive, such as a big SSD.

There are also options when it comes to synchronisation. Data may be sent in batches at particular intervals. Or there could be a hybrid approach, where some of the less important data is sent in batches, but alerts or anomalies are sent to the network in real time.

Industrial edge computing has security implications for manufacturers. Each new edge device is a potential gateway for a hacker or malicious actor. In centralised systems, there’s a principal data centre and protected perimeter – in distributed computing, dozens or hundreds of edge devices introduce new risks.

Distributed computing also helps blur the lines between IT hardware and operational technology (OT). This can help make operational technology vulnerable to attacks on the IT infrastructure, with the potential to disrupt production or even impact safety. There are also risks in terms of human factors, such as users employing weak passwords at the edge or unauthorised access by contractors.

To summarise, in distributed edge systems:

• Attack surface increases
• Segmentation of OT / IT is important to prevent attacks from spreading
• Human factors need to be considered

Let’s look at an example of edge-fog-cloud architecture for a smart factory. This includes:

1. Assets. Assets provide the data that is captured locally in the smart factory. Assets include components such as motors, conveyors, pumps, or entire CNC machine tools. These assets are connected locally using PLCs or industrial protocols
2. Edge layer. This includes industrial PCs and other local computing resources. In the smart factory, these systems are designed to be capable of withstanding local extremes of temperature or vibration. They also have enough computing power to carry out some analytics at low latency
3. Fog layer. In this smart factory, this is an onsite ruggedised server that aggregates data from all devices on the edge. It stores data for 30 days to enable buffering. It also enables management of all the devices on the edge. It features a firewall separating operational technology from IT for security
4. Cloud platform. This would typically be deployed in a system managed by AWS, Google Cloud, or Azure. This layer provides long-term data storage and dashboards for reporting. It may be integrated with enterprise resource planning systems at the manufacturer, or computerised maintenance management systems

Smart factories are likely to use architectures of this type in increasing numbers to gain the benefits of edge computing.

To learn more about the power of a world of connected devices, check out our infographic, or for a real-world example of smart factories, learn about how the IIoT interacts with the food industry.