Date centers consume from 2 to 4% of all electricity in the world, and large technological companies seek to reduce these costs. Among the possible solutions are new architectures, such as hyperconvergular infrastructure (HCI), and innovative cooling methods. However, Professor Martin Karsten from the University of Waterloo offers a much easier way to change only 30 lines of code in Linux.
Together with engineer Joe Domatly from Fastly, Professor Karsn developed a slight change in the nucleus of Linux, using the suspension suspension ( IRQ) to reduce unnecessary processor downloads and improve network traffic processing. This patch has already been included in the new version of the Linux 6.13 nucleus.
Testing showed that the change allows you to increase the network efficiency of Linux and increase the bandwidth by 45% without increasing delays. The development is based on the study of Karsten and his former student Peter Tsai, published in 2023.
“We did not add anything – we simply changed the procedure for performing operations, which significantly improved the use of the processor cache in the data centers,” Karsten explains. In his opinion, if large corporations, such as Amazon, Google and Meta, will make this decision, the saving of electricity can be calculated by gigavatt hours.
IEEE IFO Foli, Professor of Quins University in Belfast, supports such initiatives, but emphasizes that much more efforts will be used to radically reduce energy consumption. According to her, in addition to optimizing code, it is necessary to solve the problems of storage of excess data and the transition to renewable energy sources.
In Russia, Yandex also operates in this direction. She recently introduced an open source tool called Perforator, which helps to optimize the code, reducing the load on servers and infrastructure costs. Sergey Skvortsov, the Project manager of Perforator, notes that improving the software code is one of the most effective ways to reduce the power consumption of data centers. According to him, Perforator allows you to reduce the load on processors to 20% without loss of performance.
Given the fact that the data centers consume up to 4% of the global electricity, such initiatives can play a key role in increasing the stability of digital infrastructure.