Last year at this time, as part of an exercise to help my company and the EdgeX Foundry open-source project chart a course for the coming year, I wrote a post about where I saw the future of edge/IoT computing heading for 2022. You can find that post here. It’s equivocal business at best - trying to make predictions about the future – especially predictions on the future of technology. Casey Stengel (the famous American baseball manager and player) warned to “never make predictions, especially about the future.” Common sense advice to treat all predictions with certain skepticism. Having said that, I was also taught as a military officer – thanks to Dwight Eisenhower (American President and commander of the Allied Army in World War II) that “plans are worthless, but planning is everything.” Building edge/IoT solutions may not be like planning or executing a war (although some days it may feel like it), but I hope this year’s list will help you take stock of what you see, plan your 2023 solutions, and adjust according as edge technology inevitably changes. By the way, if you’d like to see how well last year’s predictions turned out, I provided a self-assessment at the bottom of this post.
Edge play time is over
Last year, I suggested that organizations were going to be transitioning from research, proof-of-concept, and pilot projects to full scale deployments. I also suggested that customers were going to look for more complete solutions than pieces/parts in order to satisfy their edge/IoT needs.
I am seeing evidence that this is already in full swing. Importantly, and more to the 2023 prediction, I see companies growing impatient with solution providers not able to provide them with solutions that are already working at scale and immediately demonstratable. Edge elements must be fully integrated into their choice of technology (hardware, sensors, devices, network, cloud providers, data visualization, analytics, security, management, etc.). Companies want edge solutions that are easily installed and even easier to own and operate. Speaking honestly, this is difficult for solution providers because no edge/IoT solution does it all (and don’t believe any company that says they do). Solution providers, like mine, are having to find the right partners and complimentary solutions, integrate like mad, and offer the “easy button” to companies wanting production ready solutions and visible ROI today (not next month or next year).
Creating edge solutions is hard - even when there are lots of fantastic technology ingredients available. But compounding the situation is that operating/owning the edge solution is even harder (and expensive). The edge is often operated by people with technology skill sets that are a fraction of what you might find in IT operation centers. People operating the edge systems are often doing so as a parttime or additional duty. That is why things have to be easy to run and work well.
Companies wanting to put edge/IoT solutions in place are making things clear to providers – research and play time is over. “We aren’t trying things anymore.” Edge solutions have to work now and they have to work at scale and they have to work such that front line workers can use them effectively.
OT Edge Security becomes a thing
I joked with many people in the industry that for the longest time, when I asked organizations about their edge security needs and requirements, the only specific detail about security that I would get in response is “keep us off the cover of the Wall Street Journal.” Organizations didn’t really know what edge/IoT security needed to do, but they were concerned about perceived threats (and the fallout of successful attacks) at the edge.
Threats at the edge are becoming more known. Requirements are becoming clearer and more specific. Companies are reading about the various successful attacks on the edge (Ring, St. Jude’s, Nortek, and Target as some examples) and they are becoming educated on what they want.
Companies are no longer under the illusion that closed loop networks are truly closed, that obfuscation is good enough protection because “this stuff is complicated”, or that “no one would bother to want to get access to this type of data” (who imagined Elon Musk’s flight plans would be of interest to anyone?). Organizations want to know how to protect all parts of the edge solution (from sensor to cloud). They also want to know how to detect when something seamy or unexpected seems to be going on.
I have seen a lot of edge/IoT security capability. Much of it originates in enterprise technology and helping to protect cloud native environments. Most of it doesn’t integrate easily or well with existing OT technology. It doesn’t operate well at the edge where it is often disconnected, operating under resource constraints, and has to deal with OT protocols and sensors. Some security startups are starting to recognize this, but these vendors will need to team up with more edge/IoT solution providers and be better integrated into the edge platforms (see Play time is over). Security vendors will begin to provide solutions that really understand edge vulnerabilities and provide some solutions that suppress OT based attacks.
Reinvention and disruption of hyperscalers
Cloud providers and the hyperscalers have tried and tried to lure all that precious edge data into the cloud where AI/ML and other analytics were to operate on it. The problem is, the vast transfer, storage, and compute charges associated with moving all that edge data to the cloud is crazy expensive. Trying to sift through all that data for nuggets of commercial value doesn’t always show an ROI – at least not yet. Companies are beginning to wake up to this reality. Google IoT Core went EOL this year. I am not predicting that more will follow that path. What I will predict is that the cloud providers and hyperscalers are going to re-invent themselves at the edge and figure out how to provide more value to companies building edge/IoT solutions. “Let us help you move all your data to our cloud” is not going to sell anymore. Organizations like AnyLog are helping us understand and build solutions that allow you to leave data at the edge (the ultimate in distributed databases) and provide a real time query mechanism to get the data from wherever it lives. No transport and central storage costs of the data beyond its origination point. Those hyperscalers that will have more success are going to be those that team with organizations that understand the edge and IoT the best – because cloud native is not edge native. Companies need more help in deploying, orchestrating, upgrading, managing, and monitoring the edge (and not as if the edge was just another set of cloud servers). Companies need more help figuring out what data to harvest and return to the enterprise or cloud (if they have to move it at all) and leave data that is chaff and noise at the edge. Companies need better visualization and operational control of the edge. Hyperscalers know how to do scale, they just need to do edge at scale and in a way that adds value and lowers cost. They can and will figure this out, but they are going to require help from organizations, people and projects that know the edge. Watch for lots of new product announcements, new partnerships and probably even some acquisitions as the hyperscalers finally take on “edge native”.
Not everything requires AI/ML
Do you remember when everyone wanted to be a part of the latest AI/ML startup? When AI/ML engineers and data scientists were going for $400K a year? When AI/ML companies advertising to turn edge data into profits were being purchased for multiples of valuation? Ok – so it’s not history – it is still happening. AI/ML is revolutionizing a lot of industries and spaces. But as with any supposed magical balm, it can be overapplied. There is a lot of edge processing going on – some of it might even require some sophisticated calculations and algorithms – but not all of it is needs costly ML models and AI engines. Simple rules engines and scripting engines can provide a lot of value at the edge – saving operational costs, improving safety and even generating new revenue. Edge solutions don’t always require advanced / complex skill sets to produce, nor do they require all sorts compute power to operate. As an example, there is a lot of sensor data that comes from a hydroponic grow bed (moisture, soil temperature, pH level, nitrate and nutrient levels and more). Growing the most crop with the fewest resources and the least amount of crop loss can be a delicate balance, but a sustainability scientist can find the right formulas and use some simple edge processing with some actuation control to manage the necessary agricultural ecosystem. No need to bring in costly (both to develop and to operate) AI/ML into the picture. To be sure, there are some edge problems well suited to AI/ML at the edge. Visual inference, for example, to do object detection, classification, etc. at the edge can be a valuable addition – especially when combined with other sensor readings for corroboration (what we call sensor fusion). But that complexity is not always needed. Companies are learning to keep it simple. There is still a lot of low hanging fruit (aka money to be found) by measuring a few edge values and automatically actuating when things get out of range. Edge solution providers that help keep it simple and find low hanging fruit at the edge, might become the new darlings of investors and companies looking to improve their company bottom lines.
Kubernetes still not the full answer, but…
Everyone’s edge is different, so Kubernetes can be used to deploy, orchestrate and manage containerized workloads at some edges. But Kubernetes does not solve all the issues around management at the edge and it struggles in resource constrained environments or environments that aren’t going to support containerized workloads. I outlined this is last year’s predictions. There have been and continue to be more CNCF efforts to extend cloud native – call them Kubernetes light - to the edge. Many of these have been attempts focused on shrinking Kubernetes at the cost of functionality. microK8s, KubeEdge, K3s are all options that have been traversing this path. But I am seeing a recognition on the part of the CNCF community that Kubernetes-light isn’t enough. I am not endorsing the referenced product, but articles like this one give me some hope that a better understanding of edge management needs via Kubernetes / CNCF is emerging. From that understanding, I am predicting in 2023 we will see the emergence of new approaches and architectures to help address edge management. Probably still fledgling efforts, but keep an eye out.
There are a lot of edge and IoT platforms, software, tools, etc. (proprietary and open source). These have emerged over the past 5-10 years of the hype cycle associated to edge and IoT computing. We’ve reached a point (through the trough of disillusionment and onto the plain of productivity) where consolidation is inevitable. I have alluded to a couple of key factors that drive this prediction. Companies want more complete solutions that just work (play time is over). Hyperscalers are going to re-enter the edge in a new way and they are going to adopt (perhaps take over) some of these products and impact the edge ecosystem considerably.
We see companies wanting to accelerate their edge/IoT solutions. Over the past few years, companies were buying AI/ML companies to gain control of the IP and the people in that space. Will they do the same to consolidate more of their holding on the edge solution space? I predict so. Edge, IoT, and OT skill sets (and IP) are not always part of an organization’s native capabilities and not easily developed. Given the current economic environment, resource availability, and the desire to own more of the capability, the consolidation of the space would follow similar changes in mobile, cloud and other such technology in the past.
Use Case Demand Changes
If you are a supplier of solutions in the edge/IoT space, you have probably been addressing use cases in the industrial space. In the predictions last year, I suggested that industrials were going to lead the way in adoption of edge/IoT solutions. Still true, but we are seeing other vertical spaces starting to really become important consumers of edge/IoT solutions. Climate change, energy shortages, health and environment concerns, people / staff shortages – all areas of need stemming from immediate global economic and geopolitical circumstances – are driving more use cases. Farmers are trying to figure out how we grow more food with few resources (water, fertilizer, people) than before. Retailers are trying to find ways to lower the need to have more clerks at the POS while still protecting against theft loss. Energy companies and battery suppliers are trying to figure out how to create and store more energy optimally in order to avoid shortages and lower the overall costs. Necessity is not only the mother of invention; its the driver we have been waiting for to spur more edge/IoT computing. Our planet is tired and changing. People are tired and opting out of past employment paradigms. Resources are becoming scarcer. Everywhere, all the time, all at once global situations are driving the need to do more with less and to be more sustainable. Technology is being asked to come to the rescue and edge/IoT solutions will be a big part of that.
My 2022 score card
If you have read this far, perhaps you want to know how I did last year before accepting anything I predict this year. Here is my self-assigned report card.
|Pervasive adoption of AI/ML at the edge||Partially true. We’ve seen more organizations trying to apply AI/ML in their overall edge solution. Not all of it has worked its way to running at the edge yet.||B|
|Hybrid edge-cloud architectures will be the norm||Part of almost every “edge” solution also has elements in the cloud. Per this year’s predictions, we have seen cloud providers and hyperscalers let us down in some ways, but that will change.||A|
|The industrial sector emerges from edge/IOT research mode||Play time is definitely over for the industrial sector. Make it hum, make it simple, make it scale is now the order of business.||A|
|Customers will demand solutions rather than pieces/parts||True, but admittedly I did not foresee just how fast organizations would require the entire edge/IoT solution to include everything – hardware, sensors, visualization, network, etc.||B|
|Realization that K8s is not enough edge management||A full realization is something organizations are still working on. Admittedly, bigger “edge” (more resources) also allows for more K8s in some cases. I don’t think we are all the way there yet and I think, per this year’s predictions, the K8s community is going to start to approach edge management in a way that is more supportive and with better solutions than shrunk down K8s.||C|
|Traditional IT hardware OEMs will need to develop edge/IOT strategies||They are trying in some cases (look at Dell’s recent announcement on Project Frontier). Others are EOL their efforts (HP). I suggested they “risk becoming irrelevant”. I think this is still a risk in some cases. My prediction was correct but not all the OEMs are developing to compete.||B|
|Digital twin standards are needed to ensure pace of innovation||I thought we’d be much further along. My biggest prediction fail and regret for the industry. Also part of the reason I suggest hyperscalers need to do more this year.||F|
|The use of digital ledger technology (i.e., block chain) will start to grow||Another fail. I still think this is coming but we aren’t there yet. Organizations are still collecting the data and making use of it. Providence of the data still to come.||F|
|Noise level for edge and 5G will continue||Nailed this one and it remains the same story.||A|
|COVID sped up challenges that were already exhibiting||Per my last prediction this year, this is just one part of the need to do more with less that is driving edge innovation.||A|
Past success or failure is probably not going to be an indicator for 2023, but at least I hope I have given you some food for thought to begin your 2023 planning. Remember, planning is everything.