The IoT in 2018: Four reasons to excited, four reasons to be worried
Few IoT applications will progress beyond mere data collection networks to become useful real-time information assets
Hardware and software vendors have made it easy to connect sensors to the cloud, but creating a useful IoT application requires so much more. Developers need to merge each layer of the IoT application architecture into a seamless whole. Ideally, developers could turn to a few standard interfaces or standards-based components to create this unified architecture. In fact, developers face a stew of standards not only at the common platform level but also associated with each target application segment (Figure 8).
Figure 8. IoT development touches on a diverse set of standards bodies addressing common platform issues such as telecom as well as those targeting vertical segments. (Source: AIOTI – the Alliance for Internet of Things Innovation)
Even if built on a suitable set of standards, IoT applications present significant architectural challenges in ensuring basic performance characteristics such as availability. With large numbers of physical devices connected through uncertain communication channels to cloud services, high-availability design is no easy task. The ability to insulate the peripheral sensor net and the cloud application from the failure of the other is not a common skill. Even less common is the ability to “design for failure” – a best practice for cloud deployments even with the kind of auto-failover and auto-scaling monitors available from the leading cloud providers.
After a sufficiently robust IoT platform hierarchy is in place, similar problems relate to the ability to turn data into useful results. It’s still no easy task to drink from the multiple firehoses of data expected in enterprise IoT applications. If developers simply aim the data streams at a corporate data lake, the enterprise loses the benefit of timely analysis. If developers incorporate streaming analytics for near real-time results without deeper analysis needed for filtering, the data could be compromised by outliers and simple stochastic variation. Along with other emerging tools for big data, machine-learning can offer a solution but the industry suffers from a shortage of data scientists able to use them effectively.
Of course, solutions for these and more pressing challenges are available. The potential stumbling block in turning data collection networks into IoT applications arises from the very novelty of massive enterprise IoT applications and correspondingly lack of widespread experience in building them.