The data explosion has arrived. In fact, the average person will have nearly 5,000 digital interactions per day by 2025, up from the 700 to 800 or so that people average today,” according to the International Data Corp. (IDC). It also forecasts that the collective sum of the world’s data will grow from 44 zettabytes (ZB) in 2020 to 175 ZB by 2025, for a compound annual growth rate of 32%. Keep in mind, these predictions were made before the pandemic, so the real amount will likely exceed these numbers.
This data tsunami has put more pressure on data center and edge compute infrastructure to be more reliable, resilient, and secure while quickly scaling to meet demand. IDC estimates 152,000 new devices will be connecting to the internet every minute by 2025, with global data traffic doubling every four years. All of this data traffic requires processing power to be closer to users, driving more advances in edge infrastructure.
And, it‘s not just personal data usage that is growing. Rapid advancements in AI, autonomous vehicles, IoT/IIoT, telehealth, analytics, and many other processes will drive data centers to be even more intelligent. Unfortunately, the broader industry has yet to achieve a fully software-defined data center. To this point, great strides have been made to virtualize servers, networks, and storage. However, the majority of data center power infrastructure still remains as dedicated capacity. That’s about to change.
Unlocking Stranded Power
When it comes to power infrastructure, data center providers, facility managers, and engineers have traditionally taken a no-risk view, typically overprovisioning power to accommodate peak demands and ensure availability. After all, downtime is very costly. However, over-allocating power can be counterintuitive when it comes to being more energy efficient and sustainable.
UPSs, energy storage, power distribution units (PDUs), static transfer switches, cooling systems, power sensors, remote power panels (RPPs), and generators all play a vital role in ensuring power is available 24/7. Nonetheless, unused capacity layered into each level of redundancy impacts efficiency and cost. For colocation providers, it puts pressure on another key metric: margin. Today’s approach is leaving quite a bit on the table.
What if the stranded capacity could be safely reallocated and quickly used in higher-priority areas while still protecting workloads? There are pools of stranded power capacity in just about every data center due to safety buffers, low utilization, and rarely used redundant infrastructure. This is especially true in colocation facilities that are at the mercy of their customers’ decisions and contractual SLAs.
Software-Defined Power
Costs can be lowered while utilization and uptime are increased through machine learning and AI to monitor, predict, and orchestrate dynamic power policies across power infrastructure components. Software-defined power (SDP) identifies, aggregates, and allocates these pools of stranded power within the data center. By doing so, additional power can be made available to racks, nodes, workloads, or circuits on demand, in real time, automatically. Simply put, SDP reallocates power from underutilized racks to other racks that need additional power, unlocking significant capacity (Figure 1).
Unlocking this stranded capacity can potentially circumvent, or at least postpone, the need to build-out additional data centers. When SDP is integrated within colocation environments, it enables providers to increase utilization of power capacity by right-sizing SLAs to customer’s needs. This results in a lower price per kilowatt for customers while still allowing providers to increase their margins. These providers can safely oversubscribe capacity while maintaining and, in many cases, increasing data center uptime performance. From a business perspective, SDP allows colocation providers the flexibility to offer a new, broader suite of products to its customers. Instead of selling just a five- or six-nines SLA, they can now offer different levels of availability at different price points for two-, three- or four-nines SLAs in the same space. By allowing software to dynamically manage the oversubscription of power capacity, providers can essentially take stranded pools of power and turn them into new revenue streams without risking or compromising uptime.
This is just the beginning for SDP. In the future, it will expand to cooling, the power grid, fuel cells, and more. As AI and machine learning advance, there are great opportunities for data centers to achieve efficiency and uptime at a lower cost and smaller carbon footprint.
