Designed to reduce the space required for repeater and termination equipment in a telephone company central office, the 19-inch rack format with rack-units of 1.75 inches was established as a standard by AT&T in 1922 — typically with 3 kW of power density. The rack format has remained a relative constant — albeit the technology mounted within it has changed considerably. Standard power densities of stayed around 3 kW to 5 kW per rack, too, until about 10 years ago.
Today, advancements in chipset technologies for both CPUs and GPUs continue to drive compute densities upwards, with the average density of more than 11 kW per rack and deployments that reach as high as 50 kW. Modern IT server architectures such as 5U, hyper-converged infrastructure, rack scale designs, and multiple other forms of high-performance computing (HPC) are rapidly becoming mainstream.
Given the increasing adoption of next-generation technologies, such as AI, ML, the IoT, blockchain, and virtual/augmented reality (VR/AR), densification in the data center will undoubtedly continue to climb. All these technologies call for considerably more compute, thus generating more heat, which in turn prompts customers to seek higher-density cooling solutions whilst maintaining — or better yet, lowering — overall cooling costs.
Meanwhile, accommodating unpredictable usage and growth models among hyperscalers and cloud and platform providers has become table stakes due to dynamic and increasing business demands — a reality that foreshadowed a digital economy thrust suddenly into overdrive in the wake of the pandemic, as we all adapted to working, studying and playing from home. As Microsoft CEO, Satya Nadella, recently stated, “We’ve seen two years’ worth of digital transformation in two months.”
Cisco forecasts hyperscale data centers will account for 55% of all data center traffic, 65% of data stored in data centers, and 69% of all data center processing power by next year.
As anyone in our industry is all too aware, it takes a substantial amount of energy to power and cool a data center. Hence, it’s good news that as demand grows among hyperscale operators and cloud providers, so do their commitments to green power deals with the objective of reducing their carbon footprints. In 2019, the four largest hyperscale operators procured 3.76 GW of renewable energy via power purchase agreements (PPAs).
But how can data center providers support their customers’ sustainability goals?
That depends on the customers and their goals. A few examples are deploying cooling technologies that reduce power and water usage; partnering with local utility providers in green energy programs; deploying renewable energy solutions near buildings to generate a portion of the power mix; purchasing bundled renewable energy credits (RECs); procuring PPAs; or strategically locating data center sites to draw renewable energy from nearby plants. The key is to be adaptive enough in your sustainability model and approach to accommodate these demands without foregoing best practices and standardization. It’s also important for data center providers to have access to long-term capital, partners with a deep understanding of the data center business, long-term infrastructure investments, and renewable assets.
Most important to the challenge of sustainability is the matter of how we design, build and operate the physical data centers. Adaptive data centers with highly efficient infrastructure will feature cooling systems that require less energyconsume less water (with the ability to run without any water as needed), and reduce total cost of ownership (TCO). Additionally, a greener approach to data center design and construction means mechanical and electrical systems are deployed only when additional capacity is needed. Scalability isn’t just good for the planet; it’s responsible business.
Future-proofing the data center doesn’t require overprovisioning and stranding resources, capacity, and capital. To the contrary, both sustainability and lower cost of infrastructure are within reach.