Automation at the Edge

7 min read

This post provides an overview of network automation and highlights how automation is facilitated in the last mile by edge computing.

There can be two perspectives on the title of this post. For those in the telco domain, it harkens to network automation at the edge. Simultaneously, for those in the IoT/edge computing world, it conjures up robots and the like operating autonomously at the enterprise edge. Both are correct and relevant because we need network automation at the network edge to help enable autonomous devices at the tip (i.e., the device edge). 5G and software-defined networks (SDNs) are the components that enable network automation in the telecom space, while smart devices, auto-deployment of apps and AI/ML (artificial intelligence/machine learning) inferencing help with automation at the enterprise edge.

As mentioned in an earlier post — “5G at the Edge” — the combination of edge computing and 5G connectivity enables near real-time automation and creates opportunities to enhance digital experiences and improve performance at the edge. The topology diagram in Figure 1 below should jog your memory. The challenge, however, is to identify which facet to automate first and how to go about it without any disruptions:

Figure 1: 5G telco network edge topology.

Figure 1: 5G telco network edge topology.

Sensors, actuators and other far edge devices, as they are sometimes referred to, generate relevant information. These edge devices may also perform first-stage analytics before sending the resulting data northbound to destinations in the network edge like the MEC (multi-access edge computing) for further processing and aggregation. However, fixed function sensors can only transmit raw data in an unprocessed state to the MEC. Remember, MEC basically combines elements of information technology and telecommunications networking.

A 5G network can be used to transmit the data directly from edge devices, through the radio heads and vRAN (Virtual Radio Access Network), to second-stage analytics running on the network edge. These analytics applications are usually containerized and therefore may run in clusters located in any part of the network edge (on the core network or other backhaul components instead of the MEC). Typically, however, they would be placed in a central location adjacent to storage — such as the MEC — in order to keep processing latencies lower.

Please make sure to check out all the installments in this series of blog posts on edge computing:

Benefits of network automation

There are many reasons for a business to implement network automation. One of the key benefits is simplifying network management, thereby reducing network failure and downtime, applying configurations consistently across the infrastructure and making scaling simpler and faster.

Network automation also enables businesses to optimize performance by providing operators with deep network analysis, monitoring and alerting. Through automation, administrators and operators are alerted on errors, resource utilization, etc., when pre-determined thresholds are breached, thereby setting them up to be in an excellent position to support their customer SLAs.

Managing the network can be very complex and repetitive. Network automation helps provide greater agility to the business by dynamically adapting the user demand, thereby using network resources only as and when needed and accelerating the roll-out of applications and services to meet the demand.

5G has facilitated virtual network functions (VNF) — especially network slicing — to fully realize the flexibility that virtualization promises. However, the underlying software applications for network functions must be architected to support any infrastructure and fully automate deployments and lifecycle events, such as service creation, transparent software upgrades, dynamic scalability and even recovery.

Network automation architecture

In the age of 5G and Industry 4.0, network servicing complexity will only increase. The heavily touted network slicing is not for the faint of heart. Each network slice is an isolated end-to-end network tailored to fulfill custom requirements specified by each service. One has to determine how the physical network should be architected so it can be sliced into virtual networks, unless the physical network has been modelled and all-consuming services have specified their goals in a machine-readable format. This can truly be realized only via automation based on expressed administrator intent and historical actions. Service providers are using automation to transform network operations by enabling existing staff to control more deployments with less time and effort, and the staff don’t need to be subject-matter experts on all of the services that they are deploying.

Zero-touch provisioning is all the rage now. Every vendor is promoting it, and every product brochure claims to offer it. While zero-touch provisioning is one of the automation outcomes and is required to make slicing sustainable in a business sense, predicting a network failure using AI and taking preemptive actions would be a significant boon and much-needed capability for network operators. The latter is commonly known as AIOps.

Keen to support AR/VR applications, an enterprise or end-user might ask for more network bandwidth. A service engineer supporting that account would consider ordering a new 5G network service instance using the service provider’s OSS/BSS dashboard. That should be the only “manual” step, because all subsequent tasks ought to be automated. That would include preparing the service instance, testing it, provisioning it, deploying it and monitoring it. And, if getting that service deployed requires rolling out a new network slice, it would be handled autonomously behind the scenes.

A proposed end-to-end network automation architecture is shown in Figure 2. The communication service providers (CSP) would operate in the two right columns, while the left column is where the edge endpoints and device management occurs:

Figure 2: Network automation reference architecture.

Figure 2: Network automation reference architecture.

Edge automation

“Automation infused with AI” or “AI-powered automation” are some common phrases you may be hearing these days. Many businesses are progressing to deploy artificial intelligence (AI)/machine learning (ML) applications at the edge to change how work gets done and build resiliency.

Intelligent automation can be served at the edge through AI/ML techniques, thereby putting real analytics within reach for everyone. Natural language processing (NLP) and computer vision are two of the emerging areas in AI that help transform process automation to intelligent automation. Using NLP, we are able to design a program that can read, understand and, subsequently, take action based on intent expressed through the normal style of human communication.

There is no industry vertical where automation's relevance at the edge is not applicable today. Let us look at some sample use cases:

  • Quality control: Automation supported by visualization algorithms can help detect defects in manufactured components on the assembly line.
  • Smart homes: Automation delivered through smart devices can cater to residents’ diverse needs, from detecting the sounds of break-ins to automatic control of electrical appliances. Most of us have seen robotic vacuum cleaners. These devices don't necessarily use cameras to guide them. Instead, they use various sensors to detect and even lidar to measure the world around them and their own progress through it. These include cliff sensors, bump sensors, wall sensors and optical encoders.
  • Industrial: Automation supported by video analytics can help detect and ensure safe factory operations by identifying and alerting hazardous conditions or unpermitted actions.
  • Healthcare: We see automation in the operating room by way of robots that assist surgeons in performing orthopedic, thoracic, urologic and other types of surgeries using minimally invasive procedures. The precision offered by these robotic-assisted surgeries benefits patients and eases the burden on surgeons and the surgical teams.

These examples are only the tip of the iceberg. A vast wealth of use cases can be successfully supported by employing the appropriate automation techniques at the edge. By gathering, processing, analyzing and interpreting the data at the device edge, businesses can gain faster insights into the conditions on hand, thereby allowing automated actions to be rolled out to ensure the business's overall safety and smooth operations.

Refer to the Industry 4.0 use case that drives home the benefit of automated inspection at the edge. From a computer vision solution that applies deep learning models that automatically detect quality defects in materials to the automatic deployment of inspection models onto the right devices at sites around the world — automation spans all facets. And then those models are autonomically maintained and updated.


Whether we call it automation 2.0, network automation, hyper-automation or process automation, the ultimate goal is to combine AI/ML technologies in order to automate, simplify, discover, design, measure and manage workflows at scale across the enterprise. Figure 3 shows the benefits of automation that are applicable to both 5G telecom networks and edge computing:

Figure 3: Benefits of automation.

Figure 3: Benefits of automation.

AI-powered automation at the edge truly amalgamates the business needs of automating processes to discover, decide and take action, while dramatically improving the response times, security and control on data due to the power of operating at the edge. Edge and automation are the new power combination and can place a business at the epicenter of opportunity to tap into the real benefits that drive tangible business results.

5G boosts the ability to get real-time information from edge devices, giving rise to a new class of applications in the industrial and civic arena. The flip side of automation is the generation and sharing of enormous amounts of data. The intersection of network and edge computing cannot be overstated. By using edge computing, service providers can potentially store data and run analytics closer to the network’s edge. It allows for services to be delivered faster and eliminates the need to use a backhaul network. Additionally, it can also be used to improve security when using a hybrid cloud solution because it avoids direct communication with the public cloud and allows for on-premises data storage.

The IBM Cloud architecture center offers up many hybrid and multicloud reference architectures, including AI frameworks. Check out the IBM Edge Computing reference architecture.

Thanks to Joe Pearson for reviewing the article.

Learn more

Related articles

Be the first to hear about news, product updates, and innovation from IBM Cloud