With connected devices through the Internet of Things generating unprecedented amounts of data, traditional, highly centralized compute architectures like cloud-first designs need reconsideration. To be able to process and act on data where it matters, while it matters, compute needs to be closer to data sources and much more intelligently distributed in nature.
Enter artificial intelligence — another buzz term often quoted but infrequently truly understood (especially in an Edge-core-cloud context). While Edge is a “place” (made real when used to harness value through Edge computing, and IoT is realistically a term used to describe a set of use cases, AI is another umbrella term, this time used to describe a specific set of capabilities for automated decision making.
AI itself is often broken into two categories — general and narrow. General AI relates more to the layman’s likely interpretation of AI (conjuring images of creepily anthropomorphized robots, often appearing before some form of apocalyptic events on the silver screen), meaning extremely sophisticated forms of intelligence that can handle a large set of tasks in an automated way. Narrow AI, on the other hand, has been designed to solve fewer, more specific tasks, in an incredibly efficient, often better-than-human, way (think a robotic arm in a factory line).
In a “Thin Edge” (closest to the data) environment and context, resource constraints (often physical factors such as device size, compute capabilities, networking/bandwidth limitations, remote nature of deployment, etc.) require creative problem solving to reduce the size of the “footprint” of algorithms running at the edge. So, we are often (at least for now) inherently thinking in terms of “Narrow AI” when we talk about Edge AI.
Actually, what we are really often talking about in an Edge environment is not the umbrella term “AI” at all, but, rather, a subset of capabilities under AI called Machine Learning, or “ML.” ML is just as it sounds — a means to train your system to learn (and, eventually, make decisions/actuate/etc.) on its own. Deep Learning, or “DL” is essentially multiple ML capabilities, scaled up to a much higher degree, and often processing different types of learning, bringing them together and cross-correlating them in extremely resource-intensive neural networks (designed to mimic a human brain, and the way we naturally build correlations and draw conclusions). You may find DL in some Thick (closer to the data center or cloud) Edge environments, but what we are often talking about when we hear the term “Edge AI” is actually “Edge ML.”
Real-world use cases
None of this is news to McLaren Racing, the division of the McLaren Group focused on winning Grand Prix races and World Championships. There are more than 200 sensors on a single Formula 1 race car, and McLaren collects about 100 gigabytes of data on each car on a race weekend. That’s a tremendous amount of data; however, the necessity of real-time data capture, processing, analytics and automation hits harder when you weigh the data volume with the fact that more than 100,000 data points are streaming from a single McLaren track car per second. In order to make crucial decisions about the cars, McLaren’s engineers need to access that data in real time, both at trackside and in mission control, to determine when to make a tire change, to evaluate the safety of the track and to see things, like a gear change, in the data before it’s actually heard on the track.
Welcome to Edge computing and ML/AI in real-world execution. With data processing and analytics at the Edge, McLaren’s crew members have faster access to the real-time information they need, rather than waiting for data to be processed centrally in a cloud and fed back out to the racing team at trackside. Oh, and by the way, these cars go to different racetracks — McLaren’s team members are able to “define their own Edge,” as they literally pick up and move their portable, ruggedized Edge data center from one track to another.
Beyond the Edge, McLaren has also perfected its Edge-Core-Cloud strategy, with historical data and deep analysis taking place back at the central McLaren Technology Center. There are over 20,000 parts in each vehicle, and approximately 70% of these parts are updated/improved/enhanced and redeployed annually. The data collected during the races themselves also feeds into McLaren’s Dell-powered racing simulator, which runs more than 300,000 simulations per second.
As Paul Brimacombe, head of enterprise architecture at the McLaren Technology Group, notes in the case study, “Data is the foundation of making the cars go faster.”
Another great example of Edge ML/AI innovation is Zenuity, a joint venture between Volvo Cars and Veoneer focused on software development for ADAS (Advanced Driver Assistance Systems) and AD (Automated Driving). Zenuity generates more than 4.4 petabytes of data per month (and that amount of data is still growing), all at the Edge, which in this case, is in the cars themselves.
Zenuity is also a great example of a well-managed Edge-Core-Cloud architecture and strategy. While all 4.4 PB are necessary within the cars, to make the real-time AI/ML-based decisions (from the multiple cameras and sensors deployed on each car) in order to reduce accidents and save lives, Zenuity saves only about half of the data (or 2.2 PB) for later analysis, historical data archiving, and more advanced, HPC-based deep learning to further refine and perfect its algorithms.
With the combination of data gathered by IoT devices, Edge computing and AI technologies, organizations can turn data into instant insights and a competitive advantage. This is all about bringing intelligent systems to the point of data capture, and gaining immediate value from data, rather than sending data to distant cloud or corporate data centers for processing.
Calvin Smith is the CTO for IoT and Industrial Edge at Dell Technologies.
To learn more
- To explore secure, scalable Edge and IoT solutions, visit the Dell Technologies “Get an edge on your IoT future” site.
- To see how Dell Technologies enables McClaren’s Edge AI, check out the Case Study.
- To see how Dell Technologies enables Zenuity’s Edge AI, check out the Case Study.
- To hear Zenuity executives explaining their well-orchestrated Edge-first AI approach, watch the Short Video.