Published in News

Underwater data centres risk sonic attacks

by on17 May 2024

Turning benign whales into terrorists

New research suggests that Underwater Data Centres (UDCs) might be at risk from acoustic attacks, which could be as simple as broadcasting a high-pitched musical note underwater.

For those not in the know, UDCs utilise the ocean's natural cooling ability, which could significantly reduce the energy usage and emissions compared to traditional data centres.

Vole tested one in 2018, where servers were submerged off the coast of Scotland. This demonstrated the potential of this technology, which has since been adopted by other firms. However, this approach also introduces specific vulnerabilities.

The University of Florida and the University of Electro-Communications in Japan have found that UDCs are vulnerable to a novel form of sonar attack due to the efficient long-range sound propagation in water.

The study showed that targeted acoustic waves can disrupt the servers underwater. Sound travels well through dense water, and attackers could induce malfunctions and crashes by targeting the resonant frequencies of hard drives. Extended exposure could cause irreversible damage to the storage devices.

Professor Sara Rampazzi, the lead researcher from the University of Florida, explained, "A denial-of-service attack could take just a few seconds, depending on the strength of the acoustic signal. However, the longer the sound is emitted, the more damage is inflicted on the computer storage device."

The research team conducted experiments on a computer server rack submerged in water, using a metal enclosure to mimic a UDC, in both a lab tank and a lake. They found that a tone of 5 kilohertz could disrupt computer operations from a distance of over six metres.

Using a commercial speaker to emit this tone at 152 decibels underwater—equivalent to 92 decibels in the air—resulted in compromised hard drives within 2.5 minutes.

However, the calls of blue whales, which can reach 180 decibels, could also theoretically pose a threat.

The team developed a machine learning algorithm capable of detecting patterns of disruption caused by attacks. Further refinement could allow networks to mitigate damage by reallocating resources before a total system failure.

Last modified on 17 May 2024
Rate this item
(2 votes)