Global internet data traffic volume peaked to 1 zettabyte for the first time in 2016. Data traffic has also risen by five-fold in the last five years, with more than 90% of the world’s data generated in the last two years alone.
With the proliferation of the smartphones and advanced multimedia capabilities, this traffic is going to increase at the same or even an increased pace. Moreover, with the almost ubiquitous nature of cellular connectivity and better last-mile connectivity approaches, we are seeing a new phenomenon — the internet of things — for the last few years. The forecast for the number of connected devices by 2020 is anywhere between 20 and 50 billion. Proliferation of these connected devices and “things” will lead to a flood of data emanating from them.
Those who understand data would vouch to the fact that the real value of data is in analytics, through which actionable intelligence can be derived and decision making processes are evolved. With the proliferation of a magnified number of connected devices emanating data, enterprises are posed with a unique set of new challenges in assimilating this data and deriving efficient data analytical mechanisms.
Data at the edge
Modern day sensors are becoming more capable. With the enhancement in technology, they are able to extract more data at an increased frequency. In effect, each sensory node is, day by day, becoming the source with a capacity of an ever-increasing data generation possibilities. Moving each of these chunks of data to the central processing machinery is cost-consuming and would need a capable infrastructure. Also, the long-held practice of bringing all the data into the central processing machinery, which is nothing but a small number of data centers, will no longer be a viable and a scalable model, considering the growth numbers charted earlier for the connected devices and their data generating capabilities.
It is also highly possible that some of the data generated from the sensors might be erroneous. Also, certain data points might be at the borderline and hence may be termed as outliers, and the business logic — or the implemented business use cases — may not be considering such data. Considering the evolution of sensor technologies and their data generation capabilities that we touched upon earlier, there will always be a case wherein the computing mechanism or the data analytics implementations will not have much use of all the data.
Considering all these cases, there is an obvious need of a data-filtering mechanism at the location of the data generator.
Why edge analytics?
Some edge devices, or nodes, may need the capability of decision making to trigger a localized action. An example would be a moisture sensor in an agricultural field which would trigger a local sprinkler based on moisture levels. Another example would be that of a store camera or a CCTV camera which instead of sending tons of data 24/7 can be programmed to hold a certain amount of intelligence that can spot anomalies and capture the relevant set of data (e.g., whenever a motion is detected using a motion sensor).
Edge analytics encompasses both the possibilities of filtering and decision making at the edge or the node. This approach enables the possibility of performing a certain amount of analytics in edge devices and thereby reducing the amount of data transfer from the edge device. Not transferring all data also brings with it the flipside of a possibility of missing something. But blindly capturing all data is also unattractive and is not a scalable solution to handling the big data deluge. The concept of edge analytics brings with it the possibility of designing an optimal model that provides the opportunity of managing the data transfer from the edge and data storage at data centers in an efficient way.
Drivers of edge analytics
Self-sufficiency: Edge analytics will prove to be extremely useful in remote operational use cases such as agriculture, solar farms, mining, drilling, etc. where the surrounding ambiance is constrained by the lack of a stable connectivity or a limited bandwidth. Edge analytics would enable a low-latency response through a localized and automated decision-making setup, regardless of the network capabilities.
Cheaper computing: Following on footsteps of Moore’s Law, computing is becoming cheaper and the real estate needed to compute is becoming smaller day by day. This is also one of the key drivers for installing edge computing devices. Depending on the setup of the connected devices, computing capabilities might be enabled in the edge device or in the gateway that connects a set of devices to the internet.
Secure data: Edge analytics will provide an opportunity for data and system architects to reduce the transfer of business-critical data payloads. Some of this critical data is consumed at the data source and suitable decisions taken thereby reducing the need for transfer of such data.
Efficiency and lower TCO: Edge analytics enables the near- and real-time analytics and thereby such models are naturally efficient. Reduced data transfer and lesser data storage need also means a lower TCO.
All IoT Agenda network contributors are responsible for the content and accuracy of their posts. Opinions are of the writers and do not necessarily convey the thoughts of IoT Agenda.