Bringing analytics to data
Many of today’s Internet of Things (IoT) solutions rely on centralized, platform-based solutions to collect, store and analyze sensor data from devices at the network edge. Platform solutions built on a cloud-centric paradigm require reliable, low-latency, high-bandwidth network connectivity—a requirement that’s clearly unsuitable for implementations where industrial assets are located remotely with limited, unreliable and expensive network connectivity, or when data volumes are immense.
This makes a compelling case for pushing more computation out to the edge of IoT. Advanced digital businesses are looking beyond simple storage and rules-based processing of sensor data at the edge. Their goal? Applying advanced machine learning and AI-based analytics to take advantage of the high-fidelity data available from devices at the edge, and then taking immediate action without the round-trip time to the cloud—bringing the analytics to the data.
What is the edge?
The edge introduces the challenge of extending the role of the platform. For edge analytics, a central cloud-based platform is still critical to manage and develop the analytics applications and models based on populations of devices, which when deployed at the edge are customized for the specific instance and scenario.
To help enable this coordination of cloud and edge, organizations need domain expertise around analytics applications, devices and field engineering. They need to combine traditional analytics with AI that can understand and adapt to the dynamic conditions in the field. Collectively, operationalizing how these domains work together in a centrally managed manner through to fluid deployment, operation and monitoring from the cloud to the edge is critical for any practical IoT solution.
Cloud is moving to the fog
We distinguish edge analytics from edge computing in the need to run a variety of machine learning, predictive and prescriptive models that require a more powerful compute environment with specific CPU, memory and storage requirements. Today, not every edge device is suitable for providing such an environment to train, run and retrain these models.
For example, traditional mobile phones can process rule-based logic, but are not suitable for retraining deep learning models. Or look at historians that specialize in data collection and its processing at the edge, but are not suitable for deploying generalized analytics and machine learning.