The Internet of Things (IoT) is currently revolutionizing entire consumer and industrial applications, through enabling them to take advantage of data and services provided by the billions of internet-connected-objects. One of the main enablers of the IoT revolution has been the integration of IoT devices with cloud computing infrastructures, which has empowered IoT applications to leverage the scalability, capacity and Quality of Service (QoS) of the cloud. Currently, all major cloud vendors provide support for hosting IoT applications and enable IoT vendors to offer their services based on pay-as-you-go modalities such as Software-as-a-Service (SaaS) and Platform-as-a-Service (PaaS).

Nevertheless, IoT/Cloud integration comes with its own limitations, which make it inappropriate for certain use cases, notably use requiring real-time field processing and control. These limitations include:

  • Inefficient use of bandwidth, as not all IoT data need to be streamed and stored in the cloud, given that only a limited subset of them has actual business value.
  • Network latency, as the interaction of IoT devices with the cloud is delayed, and therefore inappropriate for real-time applications.
  • Waste of storage, since excess data without essential business value are stored in the cloud (e.g., in cases of sensors recording values that do not change frequently).
  • Limited privacy-friendliness, as IoT/cloud deployments provide no easy way to isolate private sensitive datasets.

In order to alleviate these limitations, there is a need for moving computation closer to your users and the field, through introducing additional layers of processing between the cloud and the field.

Edge Computing Characteristics and Rising Momentum

During the last couple of years several approaches to deploying computation close to the field have been introduced, including Fog Computing, Cloudlets, and Mobile-Edge Computing. Despite their different names, they all entail the deployment of a layer of computational nodes at the very edge of the network i.e. close to the users and between the field and the cloud. Hence, we conveniently consider these approaches as different instances of the “edge computing” paradigm.

Edge computing is characterized by one or more layers of edge nodes (i.e. gateways) that are deployed between the cloud and the IoT devices, including gateways close to the field. The gateways can be of different types, ranging from embedded controllers and IoT devices with limited processing capacity, to entire clusters of computational nodes. The exact types of nodes to be deployed and used depends on the nature of the target application. In general, the proper deployment of edge nodes can provide several benefits including reduced latency for real-time applications, more efficient use of bandwidth and storage resources, enhanced scalability, reduced energy costs, improved environmental performance, as well as better opportunities for privacy control and data protection.

Edge computing is gradually becoming the preferred choice for architecting large scale IoT systems in both industrial and consumer settings including smart grids, smart manufacturing, intelligent transport, smart cities, healthcare and more. Recently, standards development organizations – such as the Industrial Internet Consortium Reference Architecture and the OpenFog consortium – have started to promote the use of edge computing as part of their reference architectures for IoT deployments. At the same time, equipment manufacturers like us are working hard to provide configurable, standards-based edge computing platforms, which can be flexibly deployed in different environments. 

Application Examples

Edge computing is mandated in the case of the following types of IoT deployments:

  • Large-scale distributed control systems, which integrate location-aware functionalities with real-time processes in order to support scalable, low-latency control operations.
  • Privacy sensitive multi user applications, which leverage the data isolation capabilities of edge computing in order to minimize transfer and processing of privacy-sensitive datasets in the cloud.
  • Mobile applications, which enable roaming users and fast moving objects (e.g., connected trains, self-driving vehicles) to interact with edge nodes at their vicinity and to benefit from access to local resources.

Two practical example applications with some of the above-listed characteristics follow:

Urban Security and Surveillance Applications

These are data-intensive applications, which perform real-time processing of multimedia streams (e.g., video) stemming from a vast amount of surveillance spots. They typically cover large areas such as metro networks, multiple office locations, metropolitan districts and more. By spawning application logic across several edge computing nodes/clouds, these applications can adaptively process video streams locally. Video is streamed to the cloud with very high frame rates whenever a potential security incident is detected, else low frame rates are used. To this end, data are processed at the edge based on advanced streaming queries, which include pattern detection capabilities on images and video. In this way, edge processing facilitates real-time processing for timely detection of security incidents, while at the same time leading to huge bandwidth and storage savings in the cloud. In such deployments it’s important for edge nodes to offer deployment flexibility in order to facilitate their installation in different environments. Likewise, it’s important to facilitate the deployment of different algorithms for detecting events in varying conditions and contexts (e.g. day/night time, face detection, image analysis etc.).

Digital Factory Automation Applications

In smart manufacturing deployments, there is often a need for providing real-time insights on production processes and their schedules. This is essential for supporting new production models such as mass-customization. Edge computing deployments in factories can provide human operators with real-time edge analytics insights about their work in a given station, in order to enable rapid validation of production processes and prompt identification of problems, while at the same time boosting their decisions about the next steps in the production process. Likewise, they can also simulate parts of the production process at the edge in order to provide insights on quality control, failure modes and maintenance operations, notably in terms of the station’s equipment.

Outlook

Edge computing has already a growing momentum, which is reflected in its market figures, industrial momentum and standards. In the coming years, the importance of edge computing will continue to rise, as it will become an essential building block for applications that will dominate the market. The latter include connected cars, self-driving vehicles, as well as the wide range of Industry 4.0 applications in manufacturing shopfloors, energy plants, oil refineries and more. Moreover, edge computing will empower entire new paradigms for IoT applications, including paradigms involving Artificial Intelligence and Smart Objects such as industrial robots, assistive robots, smart pumps, smart wearables and more. Smart objects will exhibit semi-autonomous behavior, but will be still able to interact with IoT/cloud platforms via appropriate edge nodes. Also, edge computing will enable the proliferation of applications that entail real-time actuation and control, which is one of the key value propositions of IoT. Indeed, IoT applications are not only about deriving insights and knowledge from IoT data, but also about closing the loop to the field and influencing the physical world.