Microsoft brings its experimental underwater data center to surface after two years with...

Humza

Posts: 710   +159
Staff member
Why it matters: With demand for data centers expected to rise significantly in the foreseeable future, tech firms providing cloud and distributed computing services are increasingly looking towards swift, cost-effective and environmentally friendly ways for the deployment and maintenance of their hardware. Microsoft, being one of the biggest players in this field, experimented with the idea of submerging a data center underwater as part of Project Natick, making its hardware physically inaccessible to humans in a bid to improve reliability and compare performance with on-land deployments. After two years, the 40 ft long steel vessel housing 864 servers on 12 racks has been brought up to surface, with Microsoft finding the solution to be feasible, reliable and energy-efficient.

Back in June 2018, Microsoft launched Project Natick that saw a 40 ft long steel tube filled with dry nitrogen submerged 117 ft underwater in the seafloor off Scotland's Orkney Islands. Powered by windmill energy from a nearby grid, the experiment was conducted to study the feasibility of an underwater data center that would result in lower latency for nearby coastal populations with significantly reduced deployment time over traditional installations.

After two years, the sealed container covered in a thin coat of algae, barnacles and sea anemones has now been brought up to surface, power-washed and removed from its ballast-filled triangular base and transported to the mainland for detailed analysis.

Swift ocean currents limited marine life growth on the vessel, which still required a power-wash for removing algae and barnacles

In its findings, Microsoft revealed the underwater data center to be eight times more reliable than a duplicate setup deployed on land. Helped by the chemically inert dry nitrogen, the absence of oxygen and humidity around the servers is said to have protected them from corrosion, alongside other factors such as a stable temperature and lack of vibrations and bumps caused by human activity.

These servers, a handful of which failed, are now being removed by Project Natick's researchers to send to Redmond for detailed analysis and determining the cause(s) behind their high reliability over traditional deployments.

Used to conduct research on climate change and forecast rainfall in sub-Saharan Africa and infectious bacteria, Project Natick was involved in finding ways to detect and treat cancer and also took part in Folding@Home's initiative to better understand Covid-19.

Instead of relying on Microsoft's Azure platform, this distributed computing project ran on generic servers with processing capability similar to that of "several thousand high-end personal computers."

With the experiment showing promising results, Microsoft is now looking to tap this solution for deploying its Azure cloud services that could see over a dozen vessels the size of Northern Isles linked together.

"More than half the world’s population lives within 120 miles of the coast," says Microsoft, which could benefit from low-latency internet for better web browsing, video streaming, and online gaming with underwater data centers located in their vicinity.

Permalink to story.

 

Uncle Al

Posts: 7,392   +5,830
So what would be the effect of the dry nitrogen environment in a sealed container, not submerged? A substantial difference? And would they achieve the same results if submerged in a fresh water environment?
 

Endymio

Posts: 346   +229
>> Microsoft revealed the underwater data center to be eight times more reliable than a duplicate setup deployed on land."

Considering that the hardware inside is at least 16 times more difficult to service when sealed in a tube on the bottom of the sea, I'm not sure that's a good tradeoff. Perhaps they should just use the same sealed nitrogen environment in a land-based center.
 
  • Like
Reactions: veLa and 0dium

mbrowne5061

Posts: 1,599   +905
>> Microsoft revealed the underwater data center to be eight times more reliable than a duplicate setup deployed on land."

Considering that the hardware inside is at least 16 times more difficult to service when sealed in a tube on the bottom of the sea, I'm not sure that's a good tradeoff. Perhaps they should just use the same sealed nitrogen environment in a land-based center.
"Duplicate setup" implies that dry nitrogen was used on dry land.

The difference is the fluid it was submerged in. It really cannot be understated just how much energy liquid water is capable of both storing and transferring. While the one on land was dealing with whatever the ambient air temp was doing on any particular day, the water on the bottom of the sea was probably locked in somewhere between 5-15C. My guess is that the temperature inside of the container never drifted by more than a degree, without any kind of additional energy input into the system to cool it. This alone would help to protect the electronics, since thermal cycling can do a real number on SMT components (what most electronics manufacturers are using whenever they get the chance)
 
  • Like
Reactions: Bamda

messy

Posts: 9   +11
TechSpot Elite
>> Microsoft revealed the underwater data center to be eight times more reliable than a duplicate setup deployed on land."

Considering that the hardware inside is at least 16 times more difficult to service when sealed in a tube on the bottom of the sea, I'm not sure that's a good tradeoff. Perhaps they should just use the same sealed nitrogen environment in a land-based center.
you are really missing the point. you don't service these. the hardware is commodity at a larger scale than the machine. the point is proximity to population and reduced electrical needs due to the cooling basically being free. the service is likely to pull it up and replace everything
 

Endymio

Posts: 346   +229
"Duplicate setup" implies that dry nitrogen was used on dry land.
The original MS article states just the opposite.

It really cannot be understated just how much energy liquid water is capable of both storing and transferring...My guess is that the temperature inside of the container never drifted by more than a degree...This alone would help protect the electronics.
Thermal conductivity of water compared to air is about 50x higher, yes, but you're forgetting that the components in question were surrounded by dry nitrogen. With the water outside the tube, I doubt the thermals were substantially different than any reasonably-well managed datacenter, and in fact the MS team in question credits the higher reliability not to temperature control, but rather the dry nitrogen and the lack of vibration.
 

Bulllee

Posts: 28   +14
I recall reading about this experiment when Microsoft put it on the seabed.
Seems to me if only a handful of the 864 servers failed then it does look promising.
Failsafe was the buzz word when I was learning!
 

Angga B

Posts: 85   +77
This kind of experiment always interesting. BUT what about the energy required to make that huge sealed container which can withstand water pressure in such depth without leaking, including the energy to transport and submerge/surface it. They surely need to resurface, refurbish the whole container, replace the inside equipments with newer chip tech that have higher computing power every 2-5 years, or else it would not make sense to keep them on grid afterward.
 
So what would be the effect of the dry nitrogen environment in a sealed container, not submerged? A substantial difference? And would they achieve the same results if submerged in a fresh water environment?
I suspect cooling is much more efficient in sea than on land. Heat can be dissipiated by conduction directly to the see and no air conditioning is needed.
 

Uncle Al

Posts: 7,392   +5,830
I suspect cooling is much more efficient in sea than on land. Heat can be dissipiated by conduction directly to the see and no air conditioning is needed.
The same can be achieved with ANY body of water, thus my other question about placing it in a river, lake, stream ..... of course any of these would have some (if even slight) environmental impact by creating an unnatural source of heating. FYI, the first "air conditioning" was created long before Mr. Trane had his invention. It was created by using ground water, pumped through coils to help remove heat .... happened in New Port, KY. across from Cincinnati in the "wealthy" district .....
 

mbrowne5061

Posts: 1,599   +905
The original MS article states just the opposite.

Thermal conductivity of water compared to air is about 50x higher, yes, but you're forgetting that the components in question were surrounded by dry nitrogen. With the water outside the tube, I doubt the thermals were substantially different than any reasonably-well managed datacenter, and in fact the MS team in question credits the higher reliability not to temperature control, but rather the dry nitrogen and the lack of vibration.
Active cooling - what the data center on land definitely had for the final heatsink into the environment - increases vibration. Passive cooling - what the data center in the ocean had - has no vibration beyond disks.

You're correct about the original article and the use of dry nitrogen, though they only imply it, so I can see why the wording in the Techspot article was confusing. Honestly, they should have built two, and left one on land. Anything else complicates the analysis - not like they want for data when it comes to reliability on land.
 
Interesting that no marine biologist or ecologist have chimed in on how deploying a bunch of these tubes will affect thermal rise in the ocean water and impact the eco system. One tube doesn't have much impact, but what about hundreds of them?

I can't image that this won't be a concern to ecologist and marine biologist..
 

mbrowne5061

Posts: 1,599   +905
Interesting that no marine biologist or ecologist have chimed in on how deploying a bunch of these tubes will affect thermal rise in the ocean water and impact the eco system. One tube doesn't have much impact, but what about hundreds of them?

I can't image that this won't be a concern to ecologist and marine biologist..
They're going to more concerned with fossil fuels and actual pollution. These things will act as artificial reefs - good things - as far the environmentalists are concerned.
 

summermick

Posts: 39   +45
Data centers are actually environmental-unfriendly. I have been to Microsh!t's data center in Quincy, Washington. 70-80% power consumption in that down goes to the data center's cooling system. and the data center basically generates heat and useless content, which may or may not be cloud gaming stuff. data centers have been destroyed land environments for years and now it's heading to a new territory.
 

netman

Posts: 559   +194
I am sure NSA would order lots of these cylinders to secretly control data of some population anywhere in the world...!
 

mbrowne5061

Posts: 1,599   +905
Data centers are actually environmental-unfriendly. I have been to Microsh!t's data center in Quincy, Washington. 70-80% power consumption in that down goes to the data center's cooling system. and the data center basically generates heat and useless content, which may or may not be cloud gaming stuff. data centers have been destroyed land environments for years and now it's heading to a new territory.
And that is because refrigeration systems use a ton of energy. Dropping the thing to the bottom of the ocean eliminates the need for a refrigeration system. These rely on the thermal mass of the ocean, rather than a phase change in a chiller somewhere.
 

netman

Posts: 559   +194
And that is because refrigeration systems use a ton of energy. Dropping the thing to the bottom of the ocean eliminates the need for a refrigeration system. These rely on the thermal mass of the ocean, rather than a phase change in a chiller somewhere.
The fish around vessel felt comfortable feeling the heat...! :)