In a piece posted to ‘The Conversation’ recently Paul Haskell-Dowland & Kathryn McMahon discuss the potential relocation of data centre (DC) compute functions into undersea deployments, as exemplified by Microsoft’s Project Natick.
This idea isn’t exactly new. Project Natick began in 2014 with a three-month proof of concept trial, and a two-year trial in 2018 near the Orkney Islands in Scotland confirmed the viability of underwater DCs.
The interesting points in this brief but well thought out piece are that this kind of deployment makes sense on many levels:
- It’s probably cheaper because seawater cooling removes almost all the costs of air conditioning, and machine room and DC cooling.
- It’s probably faster, by virtue of being physically closer to the consumer. Many Internet content consumers live within 100kms of the coast, so a coastal data source is likely to be significantly nearer than any other location for many, assuming data carriage is coming in by subsea cable.
- Intriguingly, it was more reliable. They argue this is because of two things. Firstly, it runs in a low-humidity nitrogen atmosphere, so there is next to no risk of fire or humidity / heat-related damage to the equipment. Secondly, because it’s opportunistically so hard to physically interact with the devices, the risk of bump-and-it’s-gone physical errors (wrong box, wrong rack, wrong time to lean on an off switch) simply isn’t there.
The authors address the primary concern of ecological risk. I don’t entirely believe putting warm objects in the water is innately beneficial. Ecologies are adapted to the local climate and changing the water temperature profile may encourage invasive species (such as Crown-of-thorns starfish) to take up residence. However, the article does show a net positive increase in spawning, feeding, and sustained existence for more species.
The other problem I see is that subsea cables typically break out into a ‘headend’, which itself then has to forward traffic into many different kinds of DCs. Presumably, you can make some kind of grand central station subsea terminus to interconnect more and more of these units, but I suspect that the class of service that it makes sense to locate at the head will tend to identify itself and exclude many other models. Including, of course, the ones where technical staff need to physically enter the room and interact with the machines!
The views expressed by the authors of this blog are their own and do not necessarily reflect the views of APNIC. Please note a Code of Conduct applies to this blog.