Success of large distributed IoT implementation depends on the foresight in careful selection of the infrastructure (tools) to design and support a robust and reliable system. Together with trusted data source(s) and  strong analytics, there should also be IoT infrastructure tools for  reduction of downtime of edge devices/sensors/sensor connectivity.

However, not many PoCs progress to a feasible production system, as the proponents fail to see the big picture. There is a tendency to connect few sensors and devices and use the raw data to show some useful analytics. The availability of Azure IoT PaaS cloud or AWS helps create a secured and connected IoT. It may be a good beginning but lack of foresight in the application and thus, selection of supporting tools or infrastructure, invariably result in challenges which become difficult to overcome in later stages.

One notable challenge is scalability of the product or system. Limitations become apparent as the number of sensors and devices exceed, (say > 250) in the IoT system. Connectivity becomes a major issue as every device has to be brought up to same firmware/ middleware version. The challenges and limitations compound as edge computing is becoming ubiquitous. One has to update the algorithms in the device as well, using Over the Top Architecture (OTA).

Then management of the Application Program Interface (API) for any device, any cloud service needs to be monitored and managed as increasing connections potentially give rise to hundreds and thousands of end point possibilities resulting in failure. This requires a robust and automated management system for API  logs using a “Watch Dog” tool.

Sensor data is useless without assured calibration and scheduled recalibration. IoT system has to manage and control the life cycle of the sensors and their calibration throughout the lifecycle of the sensor in order to produce trust-worthy data.

With so many edge devices these days , one needs automation to understand the condition of their health in-terms of memory usage, CPU temp etc. In the past this was no brainer as very little will be done inside edge.  With edge analytic it is a different story.

Azure and AWS IoT PaaS cloud architects are aware of all of the above-mentioned issues. But in their attempt to address these issues they have succeeded only to a limited extent. For example, every IoT PaaS cloud offers connectivity management. But often, that is not enough because in reality IoT system connectivity can be a mix of many connections. A competent connectivity manager must track all the connectivity levels with their Tx/Rx level, connectivity logs and must be able to build a Machine Learning based model for connectivity diagnosis and API management as well. The end result, automation and machine learning will not only be applicable to IoT application, the whole IoT infrastructure has to be managed by automation and machine learning too.