- The new function can almost miraculously reduce energy consumption in data processing centers by 30%
- The suspension of the interruption request dynamically changes the use of the processor power and can be done through the OS
- Hyperscalers will most likely become great winners, and it will be interesting to see how this affects artificial art
According to reports, data processing centers are from 2 to 4% of the total electricity consumption around the world, which, for obvious reasons, for obvious reasons seeks to reduce where it is possible.
Potential solutions include the implementation of the next generation architectures, such as hyperconvergular infrastructure (HCI) and the use of expanded cooling methods.
Professor Martin Karsten from the School of Computer Sciences School, at the University of Waterloo in Ontario, Canada, has a cheaper, simpler solution. He argues that the energy consumption of the data center of data can be reduced to 30%, simply by changing several lines of the Linux code.
Small changes, great influence
Working with Joe Domatly in Fastly, Professor Karsten developed a small, non -intelligence core change of only 30 lines of code, which uses IRQ suspension (interruption request) to reduce unnecessary breaks of the CPU and improve the processing of network traffic Linux. This setting has now been published as part of the latest Linux nucleus, the version of production 6.13.
This change in code, which is reported, increases the efficiency of the Linux network and increases the throughput to 45% without increasing delay, is based on a research article under the name “Nuclei against a user -level network: do not throw the stack with interruptions“Which Professor Karsen wrote with former student Peter Kai in 2023.
“We have not added anything,” Professor Karsten said about the change in the code. – We just rearranged what is being done when, which leads to much better use of the CPC cache in the data center. It looks like a pipe pipe in production so that you do not have people running all the time. ”
The professor believes that this small adjustment can have a huge impact. “All these large companies – Amazon, Google, Meta – use Linux to some extent, but they are very selective regarding how they decide to use it. If they decide to “switch” our method in their data processing centers, this can save gigavatt hours of energy around the world. Almost every service for the service that takes place on the Internet can be positively influenced by this. ”
AOIFE FOLYY, senior member of IEEE and professor of the School of Mechanical and Aerospace Engineering at the Queen University in Belfast, welcomes potential savings, but notes that this will require much more than just a change in a few lines of code to solve wider energy problems.
“So far, there is still a long way to go,” she says. – These objects represent enormous needs for electricity, increasing pressure on electric networks and increasing the problem of energy transitions, especially in small countries. Although this is impossible to accurately calculate, the entire ICT sector, according to estimates, is about 1.4 percent of Co₂ emissions around the world. Infrastructure and operation leaders are responsible here and should consider unnecessary waste related to data storage and adherence to the generation of energy from more renewable sources. ”
Yandex recently released an open source tool called Perforator, which uses a similar approach to Professor Karsten, helping enterprises optimize their code, reduce the load on the server and ultimately reduce energy and equipment costs.
Sergei Skvortsov, who leads the team behind the punch, told us: “This last study confirms what we believed in for a long time: code optimization is one of the most effective ways to reduce the consumption of data of the data center. The perforator helps enterprises identify and correct ineffective code, reduce the use of CPO by 20% and reduce infrastructure costs – without a victim of performance. Since data processing centers consume up to 4% of global electricity, tools such as a punch can play a decisive role in making the technical infrastructure more stable. ”