Edge computing

Edge Computing: Use Cases, Key Providers, and Implementation Options

Smart technologies — like autonomous vehicles, intelligent buildings, and IIoT or industry 4.0 manufacturing — generate so much data that it causes traffic jams enroute to the servers. The elegant solution to this challenge is shifting some tasks from powerful, but remote, data centers to smaller processors at the edge, or in direct proximity to IoT devices.

While the idea is not new, its realization has become more feasible with the arrival of high-speed 5G networks. It is expected that before long 40 percent of edge-generated data will be stored and processed locally, without the need to travel to a centralized repository.

What is edge computing

Edge computing is a distributed IT infrastructure that brings processing of raw data close to its sources, primarily — IoT sensors. This allows for assigning workloads to multiple machines, rather than relying on a single computer to deal with never-ending traffic from myriads of devices. Only actionable results are eventually transmitted to the main server, often located far away, where power and rent are cheaper.

Moving a certain portion of jobs to the periphery results in higher bandwidth and lower latency compared to frameworks built around remote centralized servers.

If translated to business terms, this means
  • better performance,
  • shorter response times,
  • real-time insights, and
  • unlimited scalability.
Besides that, edge computing allows you to occupy less cloud storage space owing to the fact that you save only the data you really need and will use.

To сrystallize the idea of edge computing, let’s compare it with closely associated concepts.

Fog computing vs edge computing vs cloud computing

Terms “fog computing,” coined by Cisco, and “edge computing” are often used interchangeably as they both involve allocating processing and analytics resources closer to the points where data is generated. The key question is: How much closer?

Fog computing or fogging happens within the local area network (LAN), somewhere between the edge and large servers. In edge computing, data is processed on the devices physically attached to the sensors.

Similar to edge and fog computing, cloud computing supports the idea of distributed data storage and processing. It replaces or complements traditional data centers, enabling scalable deployment of resources across multiple locations and providing powerful tools for analytics. Yet, cloud facilities can be hundreds or even thousands of miles away from where data is produced.

In practice, three types of computing are just different layers of a system for processing IoT data. In most cases, the layers exchange information via MQTT (message queue telemetry transport) — a lightweight IoT protocol for pub/sub communications.

Edge computing architecture

IoT system architectures that outsource some processing jobs to the periphery can be presented as a pyramid with an edge computing layer at the bottom.

edge computing architecture

How systems supporting edge computing work.

At an edge computing layer, the processing is performed on edge servers that directly interface with dozens to thousands to even millions of sensors and controllers. These servers have analytics capabilities and even run ML models to make decisions on the site in real-time. For example, they can coordinate movements of robotic arms or predict equipment breakdowns.

The edge layer also filters raw data according to predefined parameters eliminating traffic congestion on the way to the cloud. A popular instance is video recognition. Instead of moving full streams to the cloud, local devices pre-process everything that cameras "see," cut off inapplicable parts and send to the server only relevant video data.

A fog computing layer bridges the edge and the cloud. Here, fog nodes or IoT gateways execute additional filtering and analysis. The layer is capable of processing more data than edge servers. Yet, many systems do without the mediation of this kind. In other words, edge computing doesn’t need fogging while fog computing can’t substitute for edge computing.

A cloud computing layer accumulates valuable data from all edge devices and fog nodes and stores it in data warehouses. That’s where the business logic resides and Big Data analytics can be run owing to huge processing power. You can learn more about IoT cloud computing services from our article Making Sense of IoT platforms.

Roughly, edge computing can be considered as an important extension of cloud computing. Let’s look at how it works in real life.

Edge computing examples and use cases

Edge computing impacts all key industries, including manufacturing, healthcare, farming, transportation, security, and more. In particular, it drives the Internet of Medical Things (IoMT), autonomous vehicles and telematics technologies, and predictive maintenance, a proactive approach to servicing industrial machines.

Here are three most recent examples of edge computing use to inspire you.

Edge AI helps US Postal Services track missing packages

US Postal Services (USPS) delivers 7.3 billion packages a year or 231 per second. To cope with this enormous load, the company has deployed AI algorithms on its edge servers located across 195 sites. Each server has a built-in optical character recognition (OCR) functionality to analyze images from more than 1,000 mail sorting machines daily.

USPS sorting machines

Daily, each edge server processes 20 terabytes of images from more than 1000 USPS sorting machines. Source: CNN

Running locally, deep learning models categorize packages, check if the postage matches a parcel’s size, weight, and destination, and decipher barcodes, even the damaged ones. Edge intelligence also helps locate missing parcels: With AI, this takes a couple of people and just a few hours — instead of the previous several days and 8 to 10 people.

Kepler Night Nurse Edge Box increases patient safety

Kepler Vision, a Dutch medtech company, designed its Night Nurse Edge Box to keep elderly patients safe at night. The device runs Kepler software for detecting falls and physical distress and alerting staff when their help is required.

Instead of sending visual data to the cloud, Edge Boxes use computer vision locally and decide if nurses should intervene. This makes the system unaffected by the Internet connection breakdowns. Besides, it was estimated that replacing old sensors with Edge Boxes eliminates 99 percent of false alarms.

Spanish connected smart tunnel offers assistance to drivers

The Cereixal tunnel in Spain’s Galicia region leverages 5G and edge computing to capture and analyze data from tunnel sensors, cameras, and connected vehicles. Managers can remotely monitor what is happening inside the infrastructure via a dashboard.

Drivers who are moving through the tunnel receive notifications and alerts on the presence of pedestrians and emergency vehicles, possible delays because of traffic jams or accidents, weather conditions at the exit, and more. The project is supported by leading telecommunication companies Telefonica and Nokia.

Key edge computing providers

In response to a growing demand, many tech giants have launched their edge computing platforms. Key players here are
  • large cloud providers who have the core IoT software infrastructure and a wide range of related services and
  • hardware manufacturers who produce sensors, microprocessing units (MPUs), networking equipment, etc.
As a rule, the latter partner with the former to take advantage of their storage and processing capabilities. Below, we’ll look at edge offerings from four popular vendors — two from each group.

Main edge computing providers

Hardware and software offerings from main edge computing providers.

Amazon FreeRTOS and Greengrass: reduce spendings on equipment maintenance

FreeRTOS and Greengrass extend the AWS IoT platform, enabling third-party developers to program and manage edge devices qualified to work with Amazon cloud. You can get familiar with the list of Amazon partner hardware here.

A free operating system for microcontroller units (MCUs), FreeRTOS links MCU-based sensors and actuators directly to the cloud or more powerful edge devices running Greengrass.

The latter, in turn, allows you to write code and train machine learning models with AWS services and then deploy them on qualified physical platforms and IoT gateways. All communications are performed via MQTT protocol.

Amazon edge computing

Amazon edge computing offering. Source: Nordcloud

Customers and use cases

FreeRTOS facilitates building a peer-to-peer platform called SOLshare. It interconnects home solar electric systems across Bangladesh allowing them to monetize excess energy. Another application is monitoring hydraulic lifts on commercial trucks, run by Shimadzu, a manufacturer of precision instruments. This helps reduce equipment downtime and maintenance spendings.

Eco Fit, an early adopter of Greengrass, uses edge computing to analyze data from gym equipment to provide better maintenance.

Microsoft Azure Stack Edge: ensures safe ship navigation

Azure Stack Edge uses a hardware-as-a-service model to provide their customers with edge processing devices compatible with other Azure products.

Appliances ordered from the Azure portal take advantage of Microsoft’s AI and IoT services, computing, and storage capabilities. They enable you to run containerized applications and machine learning models built and trained in the Azure cloud. Devices have a local storage space and support disconnected scenarios in harsh environments.

Customers and use cases

Azure AI and edge appliances are used by JRCS, a Japanese maritime equipment manufacturer, to implement computer vision and ensure safe navigation. One more example is Olympus Medical Systems utilizing Azure equipment and AI to analyze and interpret data in real time from video cameras installed in operating rooms.

NVIDIA EGX: number one choice for smart cities

The global provider of graphic processing units (GPUs) and system-on-a-chip units (SoCs) launched its edge computing stack EGX in 2019. It is compatible with NVIDIA AI Enterprise software and integrated with OpenShift, a Kubernetes platform by RedHat. This enables companies to develop and train models in the cloud and then run and orchestrate AI deployments across NVIDIA-certified servers, produced by Dell, Cisco, Lenovo, Hewlett-Packard and other industry leaders.

EGX is also pre-connected to major IoT platforms, enabling users to manage edge computing operations via AWS Greengrass or Azure IoT Edge.

Customers and use cases

The world’s top retailer Walmart chose EGX to analyze 1.6 terabytes of data generated per second by its shops. Edge AI apps perform a large number of tasks — for example, they send alerts when you need to restock shelves or open a new checkout lane.

NVIDIA is also a popular choice for smart city solutions — like analyzing data from video cameras to optimize traffic lines, improve traffic flow, and enhance safety of pedestrians.

Cisco Edge Intelligence: prevents leaks of water, oil, and gas

US-based global leader in networking, Cisco is one of the edge computing pioneers. The company offers Edge Intelligence orchestration software that runs on its industrial gateways and services routers. It simplifies data extraction from IoT sensors, using built-in industry standard connectors. Then, the software performs real-time microprocessing of this information.

The pre-processed data can be sent to multiple cloud destinations as Cisco is pre-connected with Microsoft Azure IoT, Software AG, and Quantela, a smart city automation platform.

Cisco intelligence

How Cisco Edge Intelligence works. Source: Cisco.com

Edge Intelligence is integrated with Microsoft Visual Studio, a popular code editor most developers are familiar with. So, engineers write, test, and deploy their edge software using the convenient environment.

Customers and use cases

The Port of Rotterdam uses Cisco Edge Intelligence to analyze and visualize data from their patrol vessels. Real-time insights enable predictive maintenance and improve emergency response times. Other customers are oil and gas wells and water distribution facilities that use Cisco software to remotely control their equipment and prevent leaks and breakdowns.

IoT edge computing implementation: outsource or do it yourself?

Companies that consider utilizing edge computing have a choice: to do everything themselves or to rely on vendors like above-mentioned Cisco or NVIDIA with their end-to-end solutions.

Each option comes with its own pros and cons, and the final decision will largely depend on the company’s budget, in-house tech expertise, project scale, and other factors. Answers to the following questions can help you head for the right conclusion.

How critical is it for your business to have full control over the edge?

The DIY approach allows you to keep ownership of the edge on your side. This is a significant advantage for businesses dealing with private data that is subject to strict regulations. Healthcare organizations are among them. Companies who worry about their trade secrets and know-hows will probably also be uncomfortable putting their local processing units in the hands of third-party vendors.

However, edge infrastructure from a single provider can be quite safe in terms of data leaks. Say, Cisco edge equipment and software don’t interfere with operational data. Specific information about machines stays within the internal networks of a manufacturer. If you seek support from a vendor, discuss this point from the get-go.

Do you have sufficient internal expertise in IoT and networking?

The lack of common standards in edge computing is the main obstacle in the way of its adoption. Various devices, physical platforms, and servers may require different processing power and support different communication protocols. Сompanies without internal expertise in IoT and networking often can’t handle edge deployments and maintenance on their own.

How many locations and endpoints do you have?

Orchestration and automation is another key challenge of edge computing. The more locations and endpoints you have, the more difficult it becomes to manage their day-to-day work. When it comes to large factories with millions of sensors, you need to automate as many repetitive tasks related to the edge as possible.

Again, everything comes down to experts able to set up the automation across networks. Edge computing vendors, especially those specialized in networking equipment, have specialists who can get these processes up and running.

Who will take care of data security?

Irrespective of data protection laws, two-thirds of IT companies have serious concerns about edge computing security. A larger network of devices by its nature creates a wider surface for malicious attacks. Data breaches can happen during communication between edge computing servers, or when sensitive information is processed and stored locally.

Large tech providers typically take security concerns seriously, perform regular vulnerability assessments, update firmware and software, and quickly address issues, should they occur. If you implement the edge architecture on your own, contemplate safety precautions in advance.

The vital countermeasures to edge-related cyber threats are
  • prioritizing regular software updates to all devices to ensure that you run the bug-free version with all the most current patches and
  • performing side-channel analysis to detect unusual system behaviors (like increased power consumption) or timing delays) and thus identify installation of malicious hardware or software at the edge.
Anyway, companies rowing their own boat have to invest heavily in strategies and workforce to minimize security risks.

What analytical tasks are you going to perform?

Keep in mind, that no matter the provider, edge hardware doesn’t compare with large central servers in processing power. MPUs with limited memory and performance can’t run large libraries, heavyweight algorithms, and complex tools like Apache Spark.

So, for advanced analytical tasks, you still need a powerful cloud or on-premise processing unit. Peripheral analytics, in turn, requires lightweight algorithms supporting basic processing and machine learning. Vendors bridge periphery and the core. Their offering typically includes ready-to-use edge AI models along with the environment for edge software development, deployment, and orchestration.

How to choose the right edge computing partner

Considering the complexity of the edge infrastructure, no wonder that most companies — even tech-savvy ones — finally look for assistance from third parties. The following recommendations will help you make a safe choice.

Don’t switch providers for no particular reason. If you already use Azure cloud or Cisco equipment, it’s wise to extend your partnership to edge computing. Anyway, sticking to one vendor will save you time on staff retraining, can bring discounts for loyalty and ensure compatibility across software and hardware pieces of the puzzle.

Pay attention to real-life use cases. Edge computing is still a nascent technology, and most existing providers have been doing at it just a few years. To understand if a particular organization can be really helpful, take a look at their portfolio. Edge computing in healthcare, agriculture, manufacturing, or smart cities requires different domain and technical expertise. To be on the safe side, choose the vendor with projects relevant to your industry.

Run a prototype. Many edge computing platforms offer a free tier. Say, AWS allows you to connect three new IoT Greengrass devices monthly for nothing with a one-year trial period. Seize this opportunity to test your concept.

Take a measured approach. If you are still not sure about a particular vendor and generally about the very idea, deploy just a portion of your computing to the edge. You can scale out later when you make up your mind.