The two faces of the Internet of Things in the Data center

Posted on by

The ascension of the Internet of Things (IoT) will have two profound effects on data center operations. First, IoT techniques will continue to increase their utility for core data center operational processes, and second, broader use of IoT-enabled infrastructure will challenge data center capacity and capabilities.

Servers and their related infrastructure have been at the forefront of IoT management for a considerable period of time. In fact, a good way to understand IoT potential is to look at the capabilities of a modern managed server. These capabilities extend well beyond just a CPU, memory and input/output ports. IoT-based management components in a modern computer include items such as:

  • Temperature sensors
  • Fan speed sensors
  • Security actuators (e.g. thumbprint scanners)
  • Power sensors
  • Moisture sensors

These components also include a broad range of other sources of telemetry data. IT monitoring systems collect and aggregate this data, and when events and actions become questionable, the system automatically issues support tickets to trigger a service call. In support of this, IoT techniques continue to evolve in the data center; examples include racks in which servers are installed became available as “managed” devices years ago,  or power distribution units and uninterruptible power supplies which communicate over the network to raise alerts when power is interrupted or battery health degrades.

With the increasing heat load of modern data centers, monitoring and managing ambient temperature becomes more and more critical. Temperature and humidity sensors are deployed throughout the facility and used to enable a “smart HVAC” response that can target cooling to specific locations.

Moisture sensors trigger alerts in the case of leaks or increasing humidity, and again can trigger either an automated HVAC response or direct human intervention, while motion sensors routinely feed security systems. Smoke and heat detectors are, of course, required by regulation and are integral parts of any data center’s IoT infrastructure.

The overall power efficiency of a facility can be aggregated from the electrical consumption of both the direct computing infrastructure as well as the mechanical systems, which is then compared to industry baselines and used to drive continual improvement. Managed services providers, in particular, stand to benefit from the ongoing application of IoT management in their facilities, and its support for efficiency and economies of scale.

In many ways, the IoT is a massive expansion of these kinds of practices. The whole world becomes the data center, with sensors embedded in everyday items and industrial infrastructure. Even in the data center, IoT-based management consumes capacity and the ability to monitor heat, humidity, moisture, smoke, power and so forth requires network bandwidth, processing power and storage. When the whole world is being managed this way, IT consumption increases massively. With sensors deployed everywhere, from utility grids to medical devices to supply chains, what might be a few million data points in the modern data center quickly becomes trillions.

With these conditions, network bandwidth, compute power and storage requirements all increase exponentially. Switching and routing infrastructures will require upgrades, especially closer to the network edges where the majority of the IoT lives. Storage systems will be subject to similar loads, requiring petabytes of storage; and because data formats are so variable and logging is so write-intensive, structured relational databases are avoided in favor of NoSQL solutions better suited to log aggregation.

IoT data management practices will need a rigorous design for aggregating, archiving and purging; otherwise the costs of storage will exceed the IoT economic benefits. Expect regulators and lawyers to scrutinize IoT data from records management and litigation perspectives, while, as always, privacy advocates are already raising concerns.

Moving and storing data is pointless unless it is applied to achieving business value. This implies new and evolving analytics algorithms, and the platforms capable of running them. Such analytics may be run in batch across already stored data, or in more advanced forms would be based on complex event processing, where high volume data streams are analyzed in real time.

These kinds of applications are still cutting-edge and the subject of advanced research, including related topics, such as machine learning and neural networks. All such algorithms must be based on clear economic value, such as predicting and preventing outages, improving customer outcomes (boosting Net Promoter Scores) or increasing supply chain efficiency.

As endpoint counts spike inside and outside the data center, understanding their origins and rationalizing them to known inventories become ever more important. Blazent’s powerful analytic engine is ideally suited to manage and drive value from the Internet of Things’ accelerating growth; this is an area in which we invest continuously, and is an area to which our customers and prospects need to pay very close attention.