Analytics at the Edge: Are We There Yet?

The premise is obvious. Companies are trying to become more data-driven as they navigate an ever-changing market landscape.

They want data-driven insights in an instant. With many employing automation and machine learning, that instant is calculated in nanoseconds. This makes data latency a serious challenge. So, instead of shuttling data to a central cloud data center for analysis, it makes sense to analyze locally. We call it the edge computing or analytics at the edge.

The elusive use case

The major user benefit for edge computing is efficiency. But Kevin Ji, senior director analyst at Gartner, sees cost becoming a major roadblock to adding analytics to edge computing.

To understand the cost, we need to know that there is an edge where the data is created, which in today’s context can be a dumb sensor or an intelligent smartphone. Then there is edge computing. It is where the data is managed and analyzed. While smartphones have some capabilities, it requires a local micro data center with analytics processing capabilities.

All these mean additional costs. For advanced manufacturers, latency costs time and money. It offers a clear cut case for edge computing for these companies. Also, many plants already have local data centers inside their facilities. Re-engineering them as micro data centers is a simpler value proposition.  

For cases where immediate action needs to be taken, it also makes sense. Similar cases can involve creating services that align with end-customer behavior.

But for the vast majority of businesses, edge computing with analytics is a tall call when managing costs. “So that is why right at the start edge computing was not for analytics. Instead, it is for data management,” said Ji.

Intelligence becomes a differentiator

With remote working and smart manufacturing taking hold, advanced manufacturers are beginning to see the benefits in investing in intelligence at the edge.  

“For example, they may need to analyze sensor data to manage some machine activities while their workforce is not there. And a public cloud cannot offer the quick response they may want,” said Ji.  

“For another example, imagine a crowded area with CCTV monitoring. For the CCTV to detect a criminal and notify the police to arrest, you need local edge computing capabilities,” said Ji.

As companies see speed and efficiency as crucial to their top and bottom lines, analytics at the edge will become a focus.

Sensors at the edge are also improving. Ji noted that he talked to a manufacturer that is using smart sensors and sees a need to analyze the data quickly. Other areas include logistics where processing at the edge is needed for better service at remote locations and streamlining product innovation.

Yet, these cases are still not many. What is missing is a single case that can address an industry.

“But it's difficult to have a one-size-fits-all solution. Which is why I see edge computing driven enterprises who are profitable and are driven by product innovation rather than efficiency,” said Ji.

Cloud players bet on edge

Cloud platform players are not waiting for these clear-cut cases. In fact, they are investing in technologies for a future where analytics at the edge will become a mainstay.

Amazon Web Services has Lamda@Edge which can run code closer to the edge. Its AWS IoT Greengrass can run AWS Lamda functions and Docker containers for machine learning. Microsoft Azure allows companies to create low-latency Azure Edge Zones. Google Edge TPU is a purpose-built ASIC designed to run AI at the edge. Huawei also has its EC-IoT solution that is targeted at smart manufacturers.

Meanwhile, other players like telco equipment providers, infrastructure players like Equinix, and IT manufacturers like HPE offer similar value propositions. Many are already part of the growing Edge Computing Consortium.  

So, if the edge computing case is not clear, why are these vendors rushing to invest? Ji alluded to two issues that COVID-19 highlighted: resiliency and data sovereignty.

He noted that companies now see the importance of building resiliency into their infrastructure. Having analytics at the edge can help them overcome data-driven decision shortfalls if public cloud processing becomes unavailable or becomes too slow.

Another issue is data sovereignty. As regulators and governments get stricter on what data companies can send to cloud data centers, processing them at the edge first and anonymizing them before sending to the larger facilities for further processing makes business sense.  

Edge security is different

Another challenge is security. “Currently, the edge is inside the perimeter so security is not a worry. A lot of manufacturers are not very sensitive about edge security; they are more sensitive about the latency of the edge,” said Ji.

He however feels this will change as the adoption of edge computing increases. Companies will then need to shift their mindsets.

“The security implication often overlooked is that secure edge computing requires a secure platform to be present at the edge. In addition, edge computing requires robust and secure communication with backend systems,” said Dr. Klaus Gheri, vice president of network security at Barracuda Networks.

He noted that the platform and communication need to be hardened properly as “there is plenty of attack surface.” “On top of this, managing these platforms can become a complexity burden when deployed at scale,” he added.

The good news is that vendors are tackling these concerns. “Fortunately, there are devices that have been designed with security in mind that can protect the actual IoT environment against uncontrolled outside access and additionally provide a compute platform onto which edge compute workloads can be deployed,” said Dr. Gheri.

One example is Barracuda’s Secure Connector device, which is a certified device on the Microsoft Azure IoT catalog. “This Barracuda solution is more than just an edge computing platform; it was designed as a secure and scalable connectivity solution for industrial and commercial IoT devices,” Dr. Gheri explained.

Regardless of the edge application, Dr. Gheri advised implementing security in conjunction with a robust SD-WAN connectivity solution. “The limitation here is often not the computational overhead because of encryption but the quality of the connection especially across larger distances,” said Dr. Gheri.  

“Additionally cost may play a role as top quality grade lines like MPLS are not available or not economically viable. Again, a security device that warrants top-grade encryption and intelligent connectivity optimization can provide a solution,” he added.

Edge reality settles in

While Gartner’s Ji sees a bright future for analytics at the edge, with concepts like smart city, predictive maintenance, and intelligent building taking root, there is one factor that offers a huge hurdle: power.

“One is battery power, and the other is CPU power at the edge side,” said Ji. It is another reason analytics on the edge “is very limited at the moment.”

But the push for more adoption will come in the form of regulations. As governments ask for better resilience and data management support at the edge, cloud platform providers will need to find better solutions to manage data and find ways to address the power issues.

There are also architectural reasons why many cloud providers want analytics at the edge. First, a micro data center for edge computing can “multiply the number of connections” to a public cloud by becoming a connection point for other connections. This can increase the number of actual edge connections to a cloud platform “from 10,000 to more than a million,” said Ji.

Bandwidth can also benefit. “Since we do not want sensors to be connected 24/7 — when many sensors do not do compression — edge computing can manage the data,” said Ji. He sees this approach freeing up bandwidth and even reducing cloud-related costs.  

Whatever the reason, analytics at the cloud has gone beyond the hype cycle. It is now entering the maturity stage where companies are searching for the elusive use case. We just need to wait a little bit longer before we see a surge in edge computing adoption.