Winning with the IoT: the vitality of edge computing to the enterprise
Security
Other AI-related edge use cases expressly relate to security, quite possibly the greatest obstacle to edge computing today. Borkar mentions using neural networks to scrutinize video footage for “anomaly detection or threat detection,” which may involve image recognition to determine whether unauthorized people are in secure locations. Still, the difficulty in protecting the edge lies in the paradox of fortifying endpoint devices. Such devices are deliberately small, simple and lightweight for rapid deployment—making them vulnerable to infiltration since “the idea is to make these devices cheap and easy to deploy, [but] the cheaper and easier to deploy they are, the more difficult it is to protect them,” Wisniewski says. The researcher indicated there are two common forms of malicious behavior for endpoint devices. The first is denial of service attacks (such as the Mirai botnet attacks that compromised several online entities). The other involves stalking, in which people gain unauthorized access to smart devices to monitor or harass others.
In all cases, employing user behavior analytics to identify aberrations prior to compromises mitigates threats. The proliferation of central processing units (CPUs)—as opposed to GPUs—validates using centralized clouds for this purpose. With that model, the security burden is on the organization instead of the user, which is essential for consumer-based smart gadget adoption. “The benefit of having the centralized architecture is that it’s easier to spot patterns of malicious experimentation and abuse,” Wisniewski says. “You can potentially see the pattern of an attack occurring at scale because you’ve got millions of data points to analyze.”
There are numerous approaches to fortifying endpoint devices. The first is to ensure devices only transmit data to other authorized devices. A best practice is to use devices solely for publishing or receiving data—not both.
“Like industrial sensors, all they should do is tell you the temperature of the pipeline and do it in some cryptographically secure manner so that somebody can’t pretend to be it and tell you the wrong temperature,” Wisniewski says. “It has to have some sort of identifier so it can’t be forged, and it should be a one-way transmission.”
Certificate-based authentication methods can also help, requiring devices to have necessary certificates to transmit to or from the edge. Additionally, organizations can extend enterprise security measures (including encrypting data traffic) to gateways in lower form factor nodes to fit the platform or the device, “which is a very different take than if you start from an edge-centric, separate silo standpoint,” Norris adds.
5G implications
Impending 5G network connectivity is also projected to greatly impact edge computing efforts. The increased bandwidth and pervasiveness of 5G could prove influential in using the edge alongside centralized cloud models. According to Guard, “If you put a 5G network in place with that bandwidth, what you actually get is a network decision model now, instead of centralized or edge only.” Although such a model is currently possible, its implementation with 5G makes it even more viable for low latency, edge streaming applications.
Guard describes a use case in which endpoint hardware is placed inside locomotives for a number of functions, including “predictive maintenance, network effects and optimization.” As a result, those trains have “self-awareness,” Guard continues. “They also have network awareness. That’s a way different model than I’m going to send a stream of data back overnight of what those locomotives did today. You have no decisions in real time with that.”
Ultimately, the paradigm Guard describes leverages 5G for improved device-to-device communication. It should also affect the way organizations are able to coordinate their edge deployments to structure them around business models and customer expectations. Negahban categorizes that distinction as “the kind of orchestration ideas that are going to come out of where, if I have a fast connection to 10,000 devices right now, how can I orchestrate them in a more creative way so that my platform ultimately gives a better user experience?”
The edge imprint
In retrospect, current advancements in edge computing are making centralized cloud models more practical for the IoT. Edge platforms, analytics filtering and edge model evaluations refine the data needed for centralized aggregation, partly with AI’s pattern recognition capabilities. The edge is also critical for increased network intelligence to optimize IoT deployments. If organizations can solve the issue of security, that paradigm could make the IoT mainstream.