How to capitalize on the edge analytics development life cycle

Waiting through the "Wild West" of edge IoT is starting to pay off, with more automation software for the scale process.

IoT Word on Computer Keyboard

Image: Soulmemoria, Getty Images/iStockphoto

Edge computing and Internet of Things (IoT) devices increased by stunning numbers in 2019, when 26.66 billion IoT devices were active worldwide. In 2020, that IoT device number will grow to 30.7 billion

Small wonder there has been such focus by companies and vendors to get the edge "under control" with edge-directed networks, management, and policies, along with robust security. There have also been exploding trends in other areas of edge computing.

SEE: Special report: The rise of Industrial IoT (free PDF) (TechRepublic)

After making the initial financial investments in edge computing and IoT, management now expects real-time and near-real-time information that can tell in an eyeshot whether production is on time, if the environmentals (humidity and temperature) for fragile goods like fruits or computers are what they should be, or how traffic is flowing through urban arteries on any given Monday morning.

To fulfill these edge analytics demands, IT has to be able to manage its own production life cycle of write, test, stage, deploy, and maintain for new edge analytics applications. Many of these new analytics apps use unstructured big data.

The catch is, there are few life-cycle management frameworks orchestrated for edge computing analytics. A second complication is that the developers, who are often data scientists, come from academic environments that are research-oriented and timeline tolerant; these research environments don't have the pressures of rapid deployment and application management found in corporate settings.

SEE: Internet of Things: Progress, risks, and opportunities (free PDF) (TechRepublic)

In the corporate context, the expectation is that an edge analytics application will be able to work at scale. It's not adequate for an edge analytics app to just perform well on a small set of test data--the app also needs to demonstrate that it can scale up to the full onslaught of IoT data present in a production environment and continue to deliver accurate analytics results.

"In the past, there haven't been enough tools available to work on the logical [software] level of edge analytics development," said Nima Negahban, chief technology officer at Kinetica, which provides software that can manage the edge analytics application development life cycle. "Developers had difficulty tracking which version of their analytics applications they were running."

Negahban talked about an "inference structure" that enables software to automatically infer, track, and train analytics application features and functions, and then tie these elements into data models and feature sets (the latter consists of the structural strings and pattern recognitions that are employed in analytics searches). This infer, track, and train automation enables developers to "grow" their data models with newly learned insights from machine learning and human input, and also to scale out applications to full-scale production data test scenarios before deployment into production. During the process, you can plug in your own raw data, data modeling, and analytics training methodologies; the rest of the process is automated. 

SEE: Securing IoT in your organization: 10 best practices (free PDF) (TechRepublic)

Software can help by assisting highly skilled data scientists with the portions of the corporate analytics application development process that they are not familiar with, such as developing, refining, and version-tracking analytics applications to meet tight deadlines, and doing the same with data models and data model feature sets. If a developer ever needs to return to an earlier version of any of these artifacts, he or she can easily do so.

To capitalize on edge analytics life cycle automation, business leaders should get their minds around their edge processing and the insights they want to derive from it. They should revisit their IT infrastructures and application test beds to see where edge automation tools best fit i, and who should be using them (likely, data scientists and IT).

For those of us who have eagerly awaited analytics life cycle tools through the "Wild West" of edge IoT, the emergence of automated soft frameworks can't come soon enough.

Also see