Data centers contribute to companies' carbon footprints, but there are many ways to reduce their impact.
April is Earth Month, making it a great time to examine your impact and carbon footprint. As many businesses are looking to set or achieve goals related to reducing their carbon footprint, there's no escaping it — data centers can be a major contributor. According to Omdia, IT will be the biggest consumer of energy for businesses in many sectors, and therefore CIOs will likely be responsible for managing that energy consumption.
How do data centers contribute to carbon footprint?
In short, electricity is required to power data centers, and generating electricity from fossil fuels releases greenhouse gases which are the basis for what is termed carbon footprint. So, reducing the amount of power needed to operate data centers and sourcing that power from renewable energy are the big-picture ways to reduce a data center's footprint. With these goals in mind, let's take a closer look at the details of how that might work.
Optimize and reduce energy use at every level of the stack
Data centers and the workloads running on their servers are inherently complex. But the good news is that means there are many opportunities for optimization. From the facility level to the racks and servers to the jobs and workloads running — there are improvements to be made. Below are some of the most critical areas where energy reductions can be made and efficiencies realized:
- Move workloads to the cloud: Large cloud providers can take advantage of economies of scale like higher server utilization, efficient power management and optimized cooling. In fact, IDC predicts that cloud migrations occurring from 2021–2024 will prevent 1,014 million metric tons of CO2 from being released into the atmosphere — the equivalent of taking over 220 million passenger vehicles off the road for a year.
- Improve the efficiency of the computation: Basically, do the same computation with less energy by embedding optimization at each layer of the cloud stack. For example, you can increase utilization by multiplexing diverse workload onto a smaller set of servers, increasing their utilization and dynamically and programmatically powering down the servers that are not used. Unused server idle power can be as high as 50% of the max power draw of a server, so power management can have a big impact. It all comes down to a more efficient use of resources per computation and smart power management.
- Optimize server cooling: A significant amount of energy can be spent on auxiliary needs like cooling servers, so decreasing the portion of power that goes to these needs versus directly powering the servers themselves is one way to decrease overall energy usage. Changing the method used for cooling can help; for example, instead of using air conditioning, some data centers make use of outside air or even liquid cooling. To optimize cooling even further, it could be dynamically matched to the servers that need it most.
- Find and shut down "zombie servers": These are servers that are running but not contributing any useful function. Some estimates put their numbers as high as 30% of all virtual servers (and 25% of physical servers), which is not only an environmental cost but also a real cost in terms of licensing and software running on the servers. They can be difficult to find — one audit discovered more than 1,000 servers that hadn't even been configured with DNS software and were, therefore, invisible to the network.
- Optimize where and when workloads run to take advantage of renewable energy: One of the often-cited issues with renewable energy is that it isn't available all the time and in all places. With IT workloads, some only need to be run periodically or on an ad-hoc basis — for example, batch jobs for artificial intelligence and machine learning, modeling and simulations, reporting or ETL processes. It makes sense to match these workloads with available renewable energy by finding the locations and times of day when running these jobs could make use of these lower emissions energy sources.
- Quantify, track and measure progress: Without a reliable way to get visibility into where and how workloads are running — across the hybrid cloud environment — it's impossible to improve energy efficiency. And while, currently, most carbon footprint reporting is voluntary, it may become a requirement in the near future. That's why it's important to have a consistent way to quantify and compare energy and carbon footprint metrics across multiple clouds, cloud regions and on-premises data centers.
Putting it into practice
If some of these suggestions seem a bit daunting to you, don't worry — you're not alone. Many of these areas could require significant time and investment to optimize if starting from scratch. That's why IBM Research and IBM Automation have been working on applying AI techniques to make improving data center carbon footprints more accessible and easier to accomplish. For example, imagine if you could understand the energy mix of different regions in which your data centers reside to determine the carbon footprint variations over time. Or, what if you could understand workloads and have the ability to direct or delay the work such that it is run at the right time and right place to minimize carbon footprint, while also respecting other SLAs. These and many more are the kinds of challenging problems IBM is working to solve.
Register for the Digital Developer Conference: AIOps and Integration
If you're interested in learning more about the technologies IBM is developing in this area, you can sign up for the upcoming Digital Developer Conference: AIOps and Integration, which is being held on April 20, 2021 (just in time for Earth Day!). Check out the session entitled "The Data Center Dilemma: AI Optimization to Reduce Your Carbon Footprint" in the design track.
The conference is free and available on-demand — just register here and choose which sessions you'd like to attend. There's also a live watch party during the conference where session speakers and other developer experts stop by to hang out, do some live coding and answer your questions. We hope to see you there!
Note: IBM’s statements regarding its plans, directions and intent are subject to change or withdrawal without notice and at IBM’s sole discretion, and they should not be relied upon in making a purchasing decision.