Shrinking the carbon footprint of datacentres


The ease with which IT leaders can make use of powerful public cloud services for digitally enabled business initiatives and to propel their organisations in the artificial intelligence (AI) era means IT sustainability is rarely a top priority.

The latest data on cloud spending shows that a greater proportion of IT budgets is being spent on public cloud services. In August, analysts at Canalys reported that for the second quarter of 2024, global spending on cloud infrastructure services grew by 19% year on year to reach $78bn.

The analyst firm noted that while enterprises’ IT budgets have returned to growth, a significant portion of spending is now directed towards AI-related projects.

But Shane Herath, chair of Eco-Friendly Web Alliance, warns of the environmental impact of such tech-driven initiatives, especially as the deployment of AI technologies is highly resource intensive.

“The ICT sector as a whole generates about 3% to 4% of global emissions, and datacentres use large volumes of water for cooling,” he says. “Significant energy consumption leads to substantial carbon emissions, and the construction of new datacentres to meet AI’s growing demand exacerbates these issues.

“This contributes to water scarcity and the depletion of finite earth minerals. Moreover, the resulting electronic waste compounds the environmental burden.”

Datacentre impact of digitisation and AI

Herath and other industry experts encourage IT leaders to think about sustainability as part of their digital innovation initiatives, especially given that these initiatives increasingly require power-hungry AI servers and the cooling infrastructure needed to support them.

For the past 25 years, thanks to the economies of Moore’s Law, IT leaders have been enticed to run ever more complex datacentre applications and store greater amounts of data, with little regard to the environmental impact.

“In the new age of sustainable awareness, we now have a responsibility to react to what happened somewhere between the turn of the millennium and today,” says Mark Molyneux, EMEA chief technology officer at data security and data management company Cohesity.

He points out that corporate IT has lived through an era of cheap compute and storage, which is now nestled up against the AI renaissance. Generative AI (GenAI) and the rise of large language models (LLMs) are increasingly driving the adoption of hybrid cloud and a new and much wider distributed computing landscape. These trends, says Molyneux, mean data volumes are reaching dizzying heights.

Measuring IT consumption

In the past, the cost of IT infrastructure was directly linked to new business requirements. Providing the software and hardware to support IT-enabled business initiatives often meant purchasing additional server and storage hardware for the new enterprise applications.

Virtualisation led to server consolidation and now, in the era of cloud computing, it has become increasingly challenging for IT leaders to allocate costs accurately. FinOps aims to make these cost calculations easier, with some industry experts believing it should go further to provide metrics for IT sustainability.

Steve McDowell, chief analyst at Nand Research, believes energy-efficiency metrics are increasingly important for IT infrastructure personnel looking to optimise resources and meet sustainability goals: “Active monitoring of power metrics is an exciting new tool for [live cloud deployment users] struggling to achieve their environmental goals,” he says.

Metrics can be used to help datacentre administrators ramp up and down power and cooling to meet the environmental objectives of the business. However, James Sturrock, director of systems engineering at Nutanix, says many enterprise cloud estates are provisioned to run at a certain grade, rating and gauge level, and are not then subsequently “feathered” with enough precision engineering to make them as efficient as they should be.

Active monitoring of power metrics is an exciting new tool for [cloud customers] struggling to achieve their environmental goals
Steve McDowell, Nand Research

“If cloud application X uses database Y on cloud service Z, and it experiences peak load surges and corresponding lows that the engineering team is largely oblivious to, wouldn’t it make sense to know about these workflow spikes and troughs so energy-efficiency metrics can be more accurately mapped against IT expenditure planning?” says Sturrock.

Nutanix has been developing features in its cloud platform to provide visibility into the power consumption of a live production environment. Sturrock believes this level of visibility can help organisations improve sustainability planning, with power consumption based on measurements from the hardware in use, updated in near real time.

“Cloud engineers, developers and their corresponding operations team members will be able to visualise power metrics via a straightforward dashboard to understand energy utilisation across their deployed cloud environment,” he says.

According to Sturrock, by simplifying the process of measuring power consumption, IT departments gain detailed information on energy usage, rather than having to rely on estimations based on assumptions or “typical” consumption values.

Early attempts at incorporating sustainability measurements in FinOps have often relied on spend proxies, which lack accuracy. One example of a more detailed approach is from Anodot, a cloud cost management platform company.

Anodot recently signed a partnership agreement with Greenpixie, a cloud sustainability data company, to offer cloud cost and carbon emissions data as part of its FinOps tool. Greenpixie’s methodology – verified under ISO-14064, the carbon neutrality standard – ensures accurate cloud emissions measurements and trustworthy cloud sustainability data.

Anodot says the Greenpixie technology will be integrated with its server utilisation data. According to Anodot, cloud users’ emissions data will be calculated using the ISO-verified, cloud emissions measurement methodology developed by Greenpixie.

It hopes the partnership will enable its FinOps customers to see accurate emissions data in hourly granularity and have full visibility into the carbon and water consumption impacts of their cloud decisions.

The company says carbon has been shown to be an even more effective motivator for engineers to cut cloud waste than saving costs. However, these two motivators – cost and carbon – have not been visible in equal granularity within the FinOps tools market.

David Drai, CEO and co-founder of Anodot, expects governments, stockholders and customers to start demanding accountability for GreenOps, which will require cloud users to operate more sustainably in the cloud.

“This means that most organisations will need to measure, disclose and reduce over time the carbon footprint of their IT operations,” he says.

Drai believes that, in the next five years, the intersection of sustainability and cloud optimisation will become a practice within IT teams and the tools they use.

The role of IT consolidation

Metrics can also be used to help IT leaders consolidate IT infrastructure, avoiding the waste and costs associated with under-utilised hardware.

Cohesity’s Molyneux believes organisations should work to consolidate data onto a common platform, thereby eliminating the existence and wider proliferation of data silos. He says they can achieve this by focusing on the indexing and classification of data based on its content and value to the company, and its predefined relevant record strategy.

This process gives organisations the ability to centralise only that data which they need access to, rather than everything. It also enables a business to harness “sister” techniques in data management, including data deduplication and compression.

Molyneux claims such an approach to data management would enable an IT department to reduce data rates by almost 96% while also optimising storage resources and saving money as they boost operational efficiency from the start.

For Nutanix’s Sturrock, when organisations start to visualise real-time power usage metrics and report historical data in support of their sustainability goals, IT practitioners can strike the right balance of performance and efficient delivery of apps, data and compute.

“Technologies such as virtualisation, containers and hyperconverged infrastructure consolidate workloads onto fewer physical devices, reducing energy consumption and carbon emissions compared to traditional infrastructure,” he says.

IT sustainability and cost-efficiency

Along with consolidation, analyst Gartner urges IT leaders to increase storage utilisation to 80% and server utilisation to 65% within their datacentre operations. These measures, it says, can lead to as much as a 60% saving on costs and lower energy demands.

By extending the lifespan of servers and network devices, organisations can defer the purchase of new equipment, potentially saving up to 40% on associated costs [and reducing] e-waste and the environmental impact of manufacturing and shipping
Autumn Stanish, Gartner

“Use performance analytics tools and AIOps platforms to continuously manage the utilisation and efficiency of datacentre infrastructure,” the analyst firm recommends in its Unlock the business benefits of sustainable IT infrastructure report.

The single most effective action IT leaders can take for the environment and their budget, according to Gartner, is to defer purchasing new equipment and better manage, optimise or redeploy what they already have.

In the report, Autumn Stanish, a director analyst at Gartner, says: “By extending the lifespan of servers and network devices, organisations can defer the purchase of new equipment, potentially saving up to 40% on associated costs. This approach not only reduces e-waste and the environmental impact of manufacturing and shipping, but also directly benefits the bottom line.”

Finally, Gartner suggests that IT leaders should set lifecycle standards at the maximum duration under which they still have support from the datacentre equipment manufacturer, and then make plans to reduce lifecycles down from that only on an as-needed basis or when performance demands necessitate replacement.



Source link