Infusing AI and Infrastructure Management into Data Centers

Hybridize to get Ahead, Modernize to Keep up

All over the world, the data center industry is going through an immense transition, with several technological and business trends reshaping the way workloads are deployed and managed. These new conditions impact power, cooling, and resiliency, and traditional enterprise IT departments have to raise their game to try and match the efficiencies that hyperscale cloud providers are able to deliver.

According to this year’s 451 Research Voice of the Enterprise (VotE), Digital Pulse: Workloads and Key Projects data, 58% of enterprises are moving toward hybrid architectures, a trend that puts immense pressure on IT departments to modernize.

In particular, to compete with decreasing public cloud prices, enterprise facilities must be highly efficient, and they must comply with rising regulations around data sovereignty. Rather than fully replace legacy infrastructure, however, enterprises will most likely elect to start with improving their efficiencies, especially IT asset utilization, consolidation into larger, centralized facilities and the addition of new capacity in parts of the world not yet covered.

Automation will be a key component for all of these changes, and while software orchestration tools can manage the application layer, what about power and cooling infrastructure? Historically, data centers have been set to design points and allowed to run, but how can we make them more software-driven to dynamically match IT load?

Defining DCIM

Data center infrastructure management (DCIM) software is an integral component of this development, providing a single pane of glass and dashboard-delivered analytics for data center monitoring, asset management, and capacity planning. It helps simplify the complexity of hybrid environments, enhance productivity, and increase resource efficiency. DCIM monitors the health of the data center with power distribution metering, thresholds, and alerts, and it manages IT assets with auto-discovery tools. Some DCIMs even drill down all the way to the server and port levels, and some are able to get into capacity planning and forecasting through predictive scenario analysis. Currently, however, challenges for DCIMs include incorporating off-premise assets and the expanded cybersecurity attack surfaces that IP-connected equipment comes with.

DCIM supports all of these functions through the collection of real-time data that can be used for trending and usage analysis. Reactive DCIM improves basic monitoring with alarms or changes in performance, while proactive DCIM integrates with other systems for greater efficiency, lower risk and improved agility. Eventually, DCIM users will look to tie data into business functions like cost analysis and business planning to help drive BEV decision-making. Here, artificial intelligence (AI) can play a key role in providing greater optimization, new insights and better forecasting through automated reactions, particularly in the areas of energy optimization and data center service management. DCIM creates more software-driven data centers through improved visibility, responsiveness, efficiency, and AI can further improve provisioning, capacity planning and, in the end, provide automated management.

What AI in the Data Center Looks like Today, and Where it’s Headed

Broadly, according to this year’s VotE, AI and Machine Learning: Adoption, Drivers and Stakeholders survey, half of all enterprises have deployed AI or plan to within the next year, and this trend holds true across all verticals, especially IT and communications. Forecasting, including predictive maintenance, is a top use case, a common application that directly applies to data center management.

Application-specific requirements apply to the full stack, down to power and cooling. Wireless sensors for discrete systems used to be the norm, and now the industry has moved toward real-time alarm systems. The next step will be 3D visualization down to the rack or port, integrated with workflow management, and including both leased capacity and cloud instances (at least the data providers allow customers to see). AI will help optimize performance and forecasting across the entire set of controls. Today, there are already software tools that help with cloud orchestration, IT services management and virtual machine management, and AI will help bundle the applications with the physical layer. AI can transform DCIM data into actionable knowledge.

Datacenter-Management-as-a-Service (DMaaS) is often the starting point for AI-supported DCIM today. DMaaS combines historical data with anonymized customer data for cloud-based remote monitoring, where analysis can happen at scale, and AI can be used for anomaly detection. The long-term goal for this “second set of eyes” on data center assets will be integration with energy management, connectivity and even business costing. The more data the better, of course, but potential drawbacks to this approach include security and latency risks.

For the time being, time-sensitive monitoring and alarms will probably remain on-premises, with vendors simply offering recommendations, as operators are currently too risk adverse. On the operations side, management still generally uses legacy tools, with DCIM helping with migrations or other changes, incident management and root cause analysis. In the future, however, AI creates the possibility of the self-healing data center with no over-provisioning.

A handful of interesting examples across the industry, however, highlight how, in some ways, AI in the data center is already here today. Google DeepMind has been shown to reduce required cooling energy using historical sensor data and neural networks, while, in China, Huawei has trained its own deep neural networks to collaboratively adjust air conditioner and chiller plant controls, solving for optimal PUE and reducing associated energy costs. Huawei has also tested and deployed an AI-driven IT asset management and capacity planning product called iManager.

As DCIM becomes more and more integral to software-driven facilities, the infusion of AI will enrich optimization and, in the end, promote further automation.

Editor’s note: This is a guest post from Teddy Miller, an analyst with 451 Research.

Click the link for more information about Huawei Analyst Summit 2019.

Disclaimer: Any views and/or opinions expressed in this post by individual authors or contributors are their personal views and/or opinions and do not necessarily reflect the views and/or opinions of Huawei Technologies.