Bench Talk

(Source: Panchenko Vladimir/Shutterstock.com)

Internet of Things (IoT) devices now surpass the number of mobile phones in terms of internet-connected devices. Many of these devices rely on Wi-Fi connections, but the majority include internet connectivity with the ability to share data and usage information with trusted partners.

In IoT analytics, data analysis tools are applied to the mass of data generated from IoT devices to realize the value and benefits available within the data. That’s a fairly generic description, but the value and benefits are as diverse as the IoT applications themselves.

Let’s first explore some use cases for IoT analytics and the value it brings, and then explore some of the options for analyzing IoT data.

IoT Analytics Use Cases

Analytics can be applied directly to IoT data for the benefit of the customer or to help optimize a business’ offering within an IoT system. Consider a smart meter IoT system that measures water usage within a home. The data could be analyzed based on their city and neighborhood averages to identify whether a home has higher-than-average use and a possible water leak could exist in the home. This use model benefits the customer.

Now consider a commercial espresso machine that records usage data. The data could help identify use models of the machines that are heavily used versus infrequently used; the analysis could help in future machine designs. This use case represents actionable feedback for both marketing and design that benefits the business.

These analytics use cases are relatively trivial, but the connectivity to these sensors and systems provides a holistic benefit to customers and business alike.

As big data becomes huge data, it has to be analyzed and requires a data processing system designed with scalability in mind.

• Apache Hadoop is a big data processing system that was designed for scalability. You can run Hadoop in a single node (computer) that handles compute and storage or scale it to thousands of nodes for large-scale processing. Hadoop is designed to scale for larger data sets, with parallelization in mind.
   
• Apache Spark is another big data processing system that focuses on in-memory data processing. For this reason, it’s higher performance but can spill over into storage for partitions that don’t fit in memory.

Both Hadoop and Spark are the de facto platform for IoT processing applications. Some are even looking at analytics at the edge, using Hadoop and Spark to process data at their source.

You can find pre-packaged analytics platforms from many major companies, including IBM, Intel^®, Microsoft, and Amazon. These systems cover ingest of data, storage, and analytics. These solutions provide end-to-end management and analytics of data from collection, processing/analysis, visualization, and edge device management. Other solutions in this space include Hivemind, DeviceHive, and Splunk.

Depending on the type of data you’re processing, some storage methods and processing methods might be better suited. For example, if you’re dealing with time-series data, storage solutions can focus on the data. CrateDB is a distributed Structured Query Language database that can scale, self-heal, and process terabytes of time-series data from many disparate sensors. MongoDB and Apache Cassandra also can serve as the core of time-series data storage.

Whether you want to build your own IoT analytics solution with the help of Apache open source elements or use a cloud-based service, a variety of solutions are available that can help you find insight into your data.