Is Edge Computing all it is hyped-up to be?

Edge Computing is a topic that has generated a vast amount of digital ink since at least 2014 with vendors large and small trying to jump on a bandwagon that is often presented as "the next big thing". Even Michael Dell, one of the less flamboyant of Big Tech's grandees has predicted that "the edge will be bigger than the cloud" (speech at VMworld in 2019), and analysts have predicted some pretty eye-popping numbers by the middle of this decade (PWC forecast $13.5 billion by 2024, Gartner - $7 billion in 2021, with CAGR of 25% between 2022 and 2030). But is it real? And if so, who are going to be the big winners and losers?

Before I attempt to answer those questions, let me make a few observations.

• Edge actually became a hot topic back in the mid-90s with the move from circuit-switched to packet-switched networks and a lot of companies that weren't a big part of the telecoms ecosystem, such as Cisco (and arguably Nokia) became major players on the back of it
• At the same time, the first generation of smart devices (think Blackberry 7) and significantly more capable IoT sensors started to emerge (though the term IoT wasn't yet coined back then)
• To draw a parallel with the growth of the Internet, much of the technology has been around for a while but just as in the case of the Internet, there were a few new technologies (thanks Sir Tim Berners-Lee for www) and a critical mass of users (Metcalfe's law in action, which by my reckoning kicked in around 1995) - the growth of the Internet was largely foreseeable but largely unforeseen - just like Edge today

So what is Edge Computing and what are the technologies driving it?

Edge Computing is not the same thing as "edge" in the telecoms sense except in terms of where it happens. My working definition of "Edge Computing" is "the placement of significant compute power as close to the origin of the data being processed as is economically and technically feasible". Yes, the network is a key enabler - and a lot of Edge Computing today takes place at cell sites or in small edge DC's within the telco's footprint - but the game-changer is the ability to put high-performance, low power computing devices, typically based on ARM architecture, very close to the data source so that meaningful insights and, if needed, real-time actions can be generated with minimal dependency on the network.?

Why does this matter? Simply put, even in the most advanced economies in the G7, networks are subject to the basic laws of physics, and distance equals time - i.e. latency. Certain applications are so latency-sensitive that the sooner you can analyse and respond to an event, or an event chain, the better. Think about the systems that slow Japan's "Bullet Trains" down the moment that an earthquake is detected, or a catastrophic outage in an electricity transformer, that if not immediately detected and isolated will result in a snowballing outage that not only takes a factory offline, but actually brings down a sizeable chunk of the regional grid. For more benign examples, think about autonomous vehicles or agricultural robotics where roadside or farm-side compute can optimise equipment productivity and therefore ROI for industries like road transportation or agriculture.?

As for the technologies, there are two, and only two that really matter:

1. High performance, low power compute -?this basically means ARM cores. The ARM architecture allows the deployment of compute capacity that would previously have been possible only for high-end data centre based servers to within a few 100 metres of the data source. Complex machine learning models can be executed (if not built) close to the IoT sensor or telematic asset allowing near instantaneous response to whatever event or event-chain is detected.?
2. Near ubiquitous network access?- this doesn't mean high-bandwidth fibre networks or even 5G - with the exception of things like video streaming, most edge applications aren't particularly bandwidth-hungry. But in the past, most sensor data was simply thrown away as there was no way to process it usefully even in batch after the event, and even today, with technologies like 5G and LEO satellite-based networking it's hard to guarantee "always on". With the compute, and its associated analytics apps based very close to the data source, much more data can be analysed, and more complex patterns (event chains) can be teased out of data sets than ever before,?whether or not there is a functioning WAN. (Some estimates suggest that a good Edge Computing strategy can reduce the amount of data needing to go back to "Big Cloud" by a factor of 400:1).

So far - so good: but why no uptake??

Well, truth to tell, there is a fair amount of this already going on, but the companies who are doing a good job of deployment aren't talking about it very much as there is extreme competitive advantage in this space. As a consulting company working extensively in this area, Alan Kei Associates has found that very few of our clients are the least bit interested in sharing what they are doing, and most of our suppliers and partners tell us the same thing.?

But there are some other factors. in?

1. Lack of well-publicised use cases:?see above - most companies that are doing a good job at the edge aren't talking about it very loudly for competitive reasons.
2. Perceived architectural complexity:?Edge requires the deployment of thousands of devices and applications even within a single enterprise use case. Most "off the shelf" edge solutions are black boxes where the vendor prevents even their best customers from seeing what is inside the box - I know of one energy company that has deployed more than 50 different edge applications, each one with its own proprietary architecture and support model, which means decades of legacy maintenance debt with few, if any, economies of scale. We saw this in the early days of client/server when each vendor had their own proprietary middleware and design patterns, and until the industry settles on a more standard architectural model (and I am not talking IoT protocols here) Edge Computing will be held back from broader deployment.?
3. Failure of established OEMs to understand the nature of the opportunity:?while all the usual suspects (DellEMC, HPE, AWS, MS Azure and to a lesser extent Google) are making noises at the edge, none of them has a coherent strategy for incorporating the edge into their existing offerings. An analogy I rather like is the rings of an onion. There isn't one edge - there are many. Big on-prem data centres and hyperscale clouds are part of it - but they are at the centre of the onion. Around that core, there are multiple layers the closer you get to the source of the data. And most of the really interesting use cases are very close to the source of the data, which is not the place that most of the big OEMs are at their best.?
4. Interesting startups lack scale and traction:?there is a prevalent myth in the industry that "edge is difficult". There is a concomitant myth that it is a hardware problem. Neither of these is true. We have seen vendors offering "Edge as a Service" who neither understood "edge" nor "as a service". We have seen vendors making small but very complex boxes with vast arrays of connectivity options built in. They are all toast, or shortly will be? Why? It isn't a hardware problem anymore. And almost all hardware companies are lousy at "service".?

So where is the action? Where should I invest?os

Like it or not, the future of most enterprise-class IT is consumption-based - that is to say - "as a service". Er.... sounds like cloud to me? So where are the Edge Cloud vendors. Going back to my "onion" model, most of the Hyperscalers will sell you an "edge" solution but it is of a size and cost that you would associate with a telephone exchange. The big OEMs (or the small guys pretending to be on the path to huge scale) don't get "as a service" - i.e. cloud. In short, there are hardly any suppliers in the existing IT ecosystem who can deliver the "outer rings of the onion" in a truly "as a service model" at a price that makes any sense to potential early adopters.?

Ah, I hear you say, there are lots of Edge Data Center companies out there. And indeed there are. But a a data centre isn't a cloud. And it is only the most basic of "as a service" models. And most of them are way too large. A 1megaWatt facility is small in the grand scheme of things, but in the outer rings of the onion, a perfectly decent edge facility can be built off only a few kiloWatts of power (which, by the way, makes it easy to provide power through solar, wind and other forms of renewable energy). This is where I believe the "action" is. Small - even very small - micro-DCs that can be deployed anywhere, on- or off-grid, built from off-the-shelf components (mostly on ARM architecture) and wrapped with a layer of managed services that make it easy for the mainstream enterprise to consume. The same facilities can, of course, provide bare metal to Hyperscalers looking for hyper-local onramps to their platforms - but the future of edge is not in hyper-specialised hardware (or trimmed down DC-grade equipment). It is in managed services on a high-performance and low-power platform, preferably powered by renewable energy resources. The market will decide, as always, but I doubt that I am wrong on this one...