"Modifying some lines of code in Linux could, according to a study, save hyperscale companies billions, although I'm not entirely convinced."

A new feature has emerged with the potential to reduce energy consumption in data centers by 30%. This improvement in CPU usage is achieved through the suspension of interrupt requests (IRQ) and can be implemented via the operating system. Major players in the sector, known as hyperscalers, could be the biggest beneficiaries, raising curiosity about the impact of this solution on artificial intelligence.

Data centers account for between 2% and 4% of total electricity consumption globally, an issue that hyperscalers are seeking to minimize. Other alternatives to achieve this goal include implementing next-generation architectures such as Hyper-Converged Infrastructure (HCI) and applying advanced cooling techniques. However, Professor Martin Karsten from the School of Computer Science at the University of Waterloo in Ontario, Canada, has proposed a simpler and more cost-effective solution.

Karsten, in collaboration with Joe Damato from Fastly, has developed a small change to the Linux kernel consisting of just 30 lines of code. This tweak utilizes the suspension of interrupt requests to decrease unnecessary interruptions to the CPU and improve network traffic processing in Linux. This modification has been included in the latest version of the Linux kernel, 6.13. Reports indicate that this change not only optimizes network efficiency in Linux but also increases performance by up to 45% without affecting latency.

Professor Karsten explained that the change does not involve adding new elements but rather reorganizing existing actions to optimize the use of CPU caches in data centers. "It's like reorganizing a production line in a factory so that people aren't running back and forth," he commented.

Karsten believes that these small modifications could have a significant impact. He highlights that many large companies, such as Amazon, Google, and Meta, use Linux in various capacities, and if they decide to adopt this method in their data centers, it could result in gigawatt-hour savings globally, positively affecting every internet service request.

On her part, Aoife Foley, a senior member of IEEE and a professor at the School of Mechanical and Aerospace Engineering at Queen's University in Belfast, emphasized that while the potential savings are encouraging, changing a couple of lines of code will not solve the broader energy challenges. Foley warned that data centers generate enormous demand for electricity, putting pressure on power grids and complicating the energy transition, especially in smaller countries. The ICT sector is estimated to account for approximately 1.4% of global CO₂ emissions.

Yandex has also launched an open-source tool called Perforator, which follows a similar approach to Professor Karsten's research, helping companies optimize their code and reduce server load, resulting in lower energy and equipment costs. Sergey Skvortsov, the leader of the team behind Perforator, emphasized that optimizing code is one of the most effective ways to decrease energy consumption in data centers.