As artificial intelligence (AI), machine learning, and data applications continue to grow, the pressure they add to today’s data centres is becoming more prominent
As there is an increased volume of communication and computational demand, the traditional static network infrastructure is struggling to keep up.
A team of researchers at Technische Universität (TU) Berlin has developed a new solution. This new solution is a self-adjusting networked technology that can adapt to real-time traffic.
This new solution is part of the European Research Council-funded’ Self-Adjusting Networks’ project, which hopes to improve data centres performance while also improving energy efficiency and resource utilisation.
Rethinking network architecture
Normal data centres networks are usually inflexible. Sever racks, which are groups of high-performance computers with powerful graphics processors, are connected through fixed pathways.
These static links do not account for the fluctuating volume or direction of data traffic, resulting in inefficient bandwidth utilisation and bottlenecks.
TU Berlin’s new network model allows for the reconfiguration of connections based on real-time traffic patterns. The ideas copy traffic systems, such as the Golden Gate Bridge’s movable median barrier, which shifts lanes to match the direction of peak traffic.
Data centres can now also redirect more intelligently, shortening communication paths and improving overall systems throughput.
Using light for faster, greener connections
A big part of this breakthrough is the use of optical switches, which allow ultra-fast changes in network configuration, measured in millionths of a second. Unlike traditional electrical switches, optical technology uses light transmitted through fibre-optic cables. This allows for higher speed but also reduces energy consumption, as the need for electrical signal conversion is eliminated.
By using programmable mirrors or varying light wavelengths, optical switches can precisely direct data streams where they are needed most. This responsiveness supports the rapidly changing demands of modern applications, especially those driven by AI and big data.
Traffic management through pattern recognition
Another part of the research involves identifying patterns in data flows. Like how certain letters appear more frequently in written language, specific servers in a data centre often communicate more regularly with each other.
By recognising these predictable patterns, the network can optimise its pathways to handle frequent data exchanges more efficiently.
This method operates based on the principles of information theory, utilising mathematical models to compress and streamline recurring data streams. The result is a smarter, faster, and more resource-conscious network that aligns with the operational needs of next-generation technologies.
Research and real-world impact
Although this began as a theoretical exploration, the Self-Adjusting Networks project is now creating applications for tech giants and cloud service providers. The researchers have laid the groundwork for the future of adaptable data centre networks that are better equipped to handle the explosive growth of AI-driven workloads.
