AWS Takes IoT Forward at re:Invent 2020

Growing focus on architecture, machine learning and packaged systems

Last month, Amazon Web Services (AWS) finished off the Internet of things (IoT) sessions of its mammoth virtual cloud event, re:Invent 2020, where many major announcements gave insight into the company’s positioning in IoT.

Taking place virtually and free of charge for the first time, re:Invent 2020 attracted 570,000 registered attendees, eclipsing the 65,000 that normally visit the physical event in Las Vegas. The loss of ticket revenue dented AWS’ revenue growth slightly in the fourth quarter of 2020, but the extended reach was a great result. And the event delivered a huge swathe of product announcements: 180 of them.

To illustrate its scale in IoT, AWS reported that its systems are connected to over 500 million IoT end-points, generating more than 500 billion messages per month. Its growing ecosystem of partners has a catalogue of over 500 qualified devices and more than 300 IoT solutions. The company claims a 35% market share for FreeRTOS, its real-time operating system for lower-power devices, which means it has four times the market share of any other rival provider of real-time operating systems.

Of the 30 or so product announcements that are highly relevant for the IoT stack, several stood out.

AWS IoT Greengrass 2.0. A complete rewrite of the novel edge–cloud system widely used in IoT. The new version is modular, taking up less memory by using only the components needed for the solution, so it can be used on hardware that is more lightweight. It now also has a command line interface, so it works with Java and Python, and AWS has open-sourced the software to make it more accessible.

Amazon SageMaker. Several updates to the machine learning service include enabling easier management of machine learning models on edge devices; easier preparation of incoming data; debugging; feature management; automatic bias detection; automated workflows; automatic testing of different types of model for suitability; and pre-built models for major applications, for example, Amazon Lookout for Vision for anomaly detection.

AWS IoT Core for LoRaWAN. This enables users to connect LoRaWAN devices to the AWS IoT Core platform directly, without the need for a separate AWS LoRa Network Server. The integration also includes Amazon’s new neighbourhood area network, Amazon Sidewalk, allowing smart home devices to be used more easily on the AWS IoT Core platform.

Fleet Hub for AWS IoT Device Management. An upgrade to AWS IoT Device Management to make it easier to work with large numbers of connected devices, including aggregating metrics about them and allowing drill-down into groups of devices, as well as initiating software updates for the whole fleet.

FreeRTOS. This now has long-term maintenance and security, with guaranteed support for two years from each release. This is a good step, although long-term support in the operations technology world typically implies at least 10 years.

Amazon Monitron. A packaged system, shown below, for machine condition monitoring. Users attach the sensors to a machine to start monitoring temperature and vibration, and feed data into pre-trained models to enable predictive maintenance. The system comes with an app and is available to buy on Amazon.

AWS Panorama. A small computing appliance for detecting cameras and processing up to 20 video streams at once. It comes with pre-built machine learning models, aimed at different sectors, and integrates with other IoT and machine learning services. There’s also a software development kit for newer camera models with more processing power, deployable from Amazon SageMaker.

Together with the many other announcements, these developments signal several important directions for AWS in IoT. Firstly, a much stronger product and service portfolio for implementing and supporting edge computing, either on customers’ own premises or as an option at the edge of a telecom operator network. This explicitly recognizes the highly distributed architecture that’s now normal for IoT systems, with the need to provide similar security and management features for all levels of the stack across the distributed devices.

Secondly, AWS is pushing to have machine learning services as a normal part of the IoT stack, with some of them set up to be used even by operations engineers who lack machine learning training. This echoes one of the central themes of the whole of re:Invent 2020, which put strong emphasis on applied machine learning solutions for a range of areas including business operations, contact centres, healthcare and industrial markets.

Thirdly, the moves reveal a much higher degree of pre-integration, moving away from a component mentality. The industrial IoT world has been a complicated mix of technologies and protocols, making it hard for suppliers to offer standardized products, often forcing customers to do a lot of the system design and integration work themselves. Now, though, it’s becoming easier for customers to see the major options that are in use and to pre-integrate them, enabling a much higher level of “plug and play”.

Fourthly, AWS’ partner ecosystem continues to grow, now offering four main approaches for customers to choose from: silicon acceleration for edge devices; device qualification for edge devices and connectivity options (with strong emphasis on LoRaWAN and operators); systems integration; and specific IoT solutions, fully developed and quickly installable.

The last of these options is really important as the industrial IoT market develops; it enables smaller and medium businesses to buy and deploy IoT with more confidence. Historically these types of company have lacked the expertise to take on system development, may not have had a relationship with a suitable systems integrator, and have been poorly served by IoT suppliers.

AWS’ approach with packaged systems mimics similar moves from Intel with its Market Ready Solutions and RFP Ready Kits, and from Dell, Libelium, Qualcomm and others, to sell pre-packaged systems addressing specific market applications. This is now feasible as the major uses are better understood. Although AWS is currently focussed on serving industrial customers, it may find that it needs to provide product or system variations for other environments, like hospitals or airports. Nonetheless, the company is taking a really important step forward.

One of the strongest players in IoT for the past few years, AWS has become a central hub around which suppliers can coalesce. But industrial IoT is a highly competitive market and other major cloud players, such as Microsoft and Alibaba, have been making significant progress in serving important vertical markets and in developing new services, like large-scale digital twins. At re:Invent 2020, AWS reminded everyone that it’s strong in this area, is developing its products and services for IoT at speed, and is sharply focussed on customers’ experience in implementing and using its systems.

A benefit of IoT deployment is that users can score even larger efficiency gains from an IoT system by expanding the scope of the system from an individual player, possibly to include a whole supply chain working from a common data framework. AWS is taking part in some projects where its customers are trying to do this, for example, Carrier with the cold chain for food and medicine. However, it’s not yet positioning itself strongly for the task, unlike Microsoft, which is increasingly taking a sector-level approach.

The industrial world is a huge and a relatively new opportunity for players such as AWS and Microsoft, which were historically focussed on IT. The market is still in its early days of development, but will heat up over the coming years as software giants increasingly make a play for it.