Edge is the New Cloud

Edge is the New Cloud

No, the edge is not going to kill the cloud. In fact, FogHorn recently did a major partnership announcement with Google, and we have strong relationships with other major cloud providers globally. So, what do I mean by “Edge is the New Cloud?” I mean that the while the cloud has brought massive constructive disruption to a broad range of industries, the edge is now the next big lever for digital transformation of the internet of things (IoT), in particular the industrial IoT (IIoT). The reasons are clear. There is simply too much data to move back and forth, security is a growing concern, and more and more applications require extremely low latency to be effective (<1 millisecond in some cases). Further, executing machine learning (ML) models and artificial intelligence (AI) at the edge generates higher quality predictive insights, delivering greater operating efficiencies including uptime, yield, and energy savings.

According to Gartner, by 2022, as a result of digital business projects, 75% of enterprise-generated data will be created and processed outside the traditional, centralized data center or cloud, up from less than 10% in 2018*.

IIoT is a Gold Mine for Industrial Companies – With an Intelligent Edge

Accenture estimates the IIoT could add $14.2T to the economy by 2020. McKinsey says the economic impact of smart factories could be between $1.2 and $3.7 trillion by 2025. A recent U.S. Department of Commerce survey of U.S. manufacturers and smart manufacturing vendors suggests $57 billion dollars in annual cost reductions. But none of this can happen without real-time streaming analytics, machine learning, and industrial artificial intelligence (AI) at the edge to address the data volume, security, latency, and predictive insights considerations mentioned earlier.

So, what is “real edge” intelligence and what does it contribute to IIoT Digital Transformation?

First let’s define “real edge” intelligence. In general terms, it is purpose-built software designed to be deployed as close to the industrial data source as possible. In some cases, this might be a manufacturer’s data center server, but in many cases, it is highly compute-constrained devices like PLCs, motion sensor kits, control systems, IoT gateways, etc. Raspberry Pi and Intel Up2 - based devices for example. But in specific terms, “real edge” intelligence starts with software that offers a hyper-efficient complex event processor (CEP) that cleanses, normalizes, filters, contextualizes and aligns “dirty” or raw streaming industrial data as it’s produced, and also handles the data post-processing. In addition, a “real edge” solution includes integrated ML and AI capabilities, all deployable into the smallest (and largest) compute footprints. By relocating the data pre and post processing functions from the ML model to the CEP, the model can be reduced by 80+%, and becomes deployable at the edge. This enables real-time, actionable analytics on site at the industrial edge, with a user experience and alerts optimized for fast remediation by operational technology (OT) personnel. It also prepares the data for optimal ML/AI performance, and enables it to be deployed close to the data source, generating the highest quality predictive insights to drive asset performance and process improvements.

Specifically, “real edge” intelligence starts with software that offers a hyper-efficient complex event processor (CEP) that conditions “dirty” industrial data as it’s produced, and handles post processing functions, shrinking ML models by 80+% so they can be pushed to the edge.

 To explain the value of “real edge” intelligence, it’s best to consider eight elements. Yes, eight but I promise to be concise. Here we go.

1)    Real-time streaming analytics, close to the data source. CEP-based, real-time analytics are ideal for industrial applications requiring low latency, and results in greater efficiencies in things like uptime, yield and energy savings. It also provides much higher data fidelity (see figure 1) than sampled batch processing. But beware the “fake edge”. Without a CEP, latency is higher, the data remains “dirty” making analytics much less accurate, and ML models are significantly compromised. A great first question to ask your potential edge intelligence software vendor. “Do you have a CEP?”

Figure 1: Data fidelity analyzing real-time streaming data versus batch sampling

2)    Iterative Edge to Cloud machine learning. Edge devices, generating continuous inferencing on live streaming industrial data (including audio and video) regularly send insights back to the cloud. These edge insights enhance the model, significant improving its predictive capabilities. The tuned models are then pushed back to the end in a constant closed loop, reacting quickly to changing conditions, and generating much higher quality predictive insights to improve asset performance and process improvements. (See figure 2)

Figure 2: Edge to Cloud Closed Loop ML

3)    Taps into OT tribal knowledge. Edge intelligence should be developed with the OT teams in mind. This allows the quick translation of operator domain expertise into analytic expressions and ML models. This is much less expensive and faster than PLC reprogramming and avoids costly, cloud-based ML exercises.

4)    Radically lower data persistence and transport requirements. These is simply too much data being produced at the industrial edge to send it all to the cloud. Processing live data at the source reduces data network and storage resource needs and can reduce cloud storage and communications costs by 100-1000x.

5)    Enhances security posture. Edge intelligence eliminates the need to transmit sensitive OT data across networks. Some environments, for security reasons, are not allowed to be connected to the Internet at all. This also reduces security infrastructure, risk mitigation, and regulatory compliance costs.

6)    Cloud agnostic flexibility. The right edge intelligence solution will be cloud agnostic to avoid any kind of lock in and facilitate multi/hybrid cloud strategies. Obviously, a byproduct of this is more bargaining power and reduced sourcing costs. This is especially critical in the early days of IIoT pricing given its complexity and heavy variance use case by use case.

7)    Leverages small footprint edge computing and controller hardware. Edge intelligence software should be able to run on industrial control systems and other highly constrained edge computing devices. This minimizes investments in heavy compute or new industrial control systems hardware.

8)    Subscription, not consumption, based pricing. Subscription-based edge intelligence software pricing makes it much easier to project scaling costs after initial proof of concepts (PoCs). The results in much more controllable / predictable operating costs and is radically cheaper for data intensive applications.

So, there you have it. Edge is the new Cloud and will be the key enabler to unlock the trillions of dollars of value creation for IIoT and move many PoCs to production smoothly and with improved technical and business outcomes. FogHorn is the established leader in edge intelligence, and we look forward to continuing to work closely with our world class roster of clients, partners and investors.

As always, let me know what you think.

Keith

*Gartner, Top 10 Strategic Technology Trends for 2018: Cloud to the Edge, Published: March 8, 2018