A Lego Approach to Edge Computing in IoT

Intel assembles the blocks for IoT edge applications so you don’t have to

The enterprise market for the Internet of things (IoT) is well known for its highly fragmented and diverse supplier landscape. This typically means that design, architecting and development of systems is a significant task, and one that has historically slowed the rate of adoption in the market.

Intel, Dell, Libelium and others have started to address this challenge for end customers and systems integrators by developing finished IoT systems. In Intel’s case, they’re known as Market Ready Solutions, serving the most common applications and industry sectors (see Intel’s Systematic Effort to Scale Industrial IoT). This approach has helped to speed up growth in the market, and is now widely copied by other IoT players.

But there are still thousands of suppliers developing and building industrial systems using IoT that face the full fragmentation of the supply chain. Can anything be done to help make their task less complex?

The automotive industry faced a similar challenge in the 1960s, as it saw demand for cars grow steeply, with a ballooning number of models. Car-makers standardized on a few chassis, engine and drivetrain “platforms”, and then used combinations of these with different car bodies to produce a broader range of vehicles.

Another useful analogy is Lego sets covering themes such as Bionicle, Star Wars, Harry Potter and others. The design, piece selection and testing are all done by Lego, based on its own common components; customers just need to assemble the kit.

Intel, with more than 50 of its IoT partners, is now taking a similar approach with the software and machine learning aspects of edge computing in IoT. Its focus is on software stacks and components to serve different uses, made available through the company’s Edge Software Hub, which launched in May 2020 and aims to become a one-stop shop for edge solutions.

Intel currently has more than 20 preconfigured and pretested full-reference implementations of software stacks for various vertical markets. Equivalent to Lego kits, they span a range of common applications including automated checkout for retail; using machine learning to analyse the very large image files from pathology labs, and optimizing the way they’re uploaded over multiple networks; monitoring activity such as Covid-19 protocols in a store aisle; traffic management at road intersections; pipelines for connecting IoT devices to cloud services such as Amazon Web Services (AWS) and Microsoft Azure; defect classification for textiles; and monitoring and control of oil and gas wells with a modern software architecture.

The common components and sub-assemblies of these edge computing solutions are also available to download as preconfigured and tested packages that customers can use to develop custom systems. Packages cover computer vision (Edge Insights for Vision), video and time series analytics for industrial usage (Edge Insights for Industrial), edge applications with network optimization (Converged Edge Insights) and real-time computing for industrial controls (Edge Controls for Industrial).

In turn, these offerings are built on several core software platforms from Intel, its partners and the open-source community. They include Intel’s OpenVINO toolkit for computer vision, the OpenNESS framework for multi-access edge computing in telecom networks, and the Linux Foundation’s EdgeX Foundry framework for edge computing in industrial gateways.

The software packages are designed to run on Ubuntu and Red Hat Linux. Each of the frameworks and underlying components has a development road map, with Intel running a quarterly update cycle for the packages. It’s up to developers or systems integrators to install updates.

Beyond the software solutions, Intel is also running a qualification and testing programme for edge computing hardware as part of Edge Software Hub, as well as a page listing recommended hardware, enabling buyers to focus on devices that are already certified for each software package. There are currently almost 70 different hardware designs validated for Intel’s software packages.

Intel has other major initiatives for edge computing, such as its DevCloud for the Edge, which allows customers to try different machine learning algorithms on various types of processor before committing to a hardware option. A natural extension of this service could be to let users try out algorithms in the context of a larger software package, as well as the processor, and to evaluate the overall system more fully.

Interestingly, while Intel has been building up its Edge Software Hub, the broader industry is developing in two complementary directions. Firstly, there’s a strong move to adopt app stores for distributing and updating industrial IoT software and algorithms, with IoT app stores available from AWS, Canonical, Microsoft, Siemens and others. Intel is clear that it’s focussing on use cases and working with partners by cross-listing relevant products in each other’s stores, rather than competing against them.

The second direction is a growing focus on vertical markets and their specific needs, with important recent launches of vertical clouds from AWS and Microsoft. Intel, with its Market Ready Solutions, has been systematically looking at differences in usage scenarios in individual industries. For example, monitoring movement of people in a retail setting isn’t exactly the same as monitoring people at a road junction. The differences may be technical, such as how the camera is connected, or may be more subtle and focus mostly on using a different machine learning algorithm.

The disciplined approach of Intel’s Market Ready Solutions will be equally relevant with the expansion of uses that its Edge Software Hub addresses.

But, importantly, Intel has for some time been deeply involved in the development of new industry-led efforts in edge computing. These include the Open Retail Initiative and the Open Process Automation Forum for industrial control systems, as well as 10 or so others for power grids, manufacturing, plus oil and gas. The experience it has gained from these initiatives should help Intel to make any necessary tweaks to its offerings for specific sectors fairly quickly, enabling the company to broaden out the range of uses and sectors it serves.

The car industry and Lego examples show that an approach based on platforms or sets enables significant economies of scope and scale, even if the development of such platforms absorbs a lot of time and cost. Companies can then use a relatively small number of major components to produce a larger set of end products.

These software platforms also enable much more rapid development of end products, which will be important in IoT, as companies look to digitize their operations to make themselves more responsive and competitive.