Microsoft submerges a data center off the coast of Scotland

Greg S

Posts: 1,607   +442
Forward-looking: Rapidly deployable data centers with low latency access for the majority of the population could become a reality very quickly if Microsoft's sunken tube of technology proves to be successful.

Project Natick began several years ago as a moonshot at Microsoft to build an underwater data center. After initial research and development phases, a three-month trial was conducted in the Pacific. Moving forward to present day, a submersible data center capsule has been deployed 120 miles off the coast of Scotland.

Inside the 40-foot long tube lies 12 racks filled with a total of 864 servers. Sea water is pumped through radiators in the back of each rack to keep everything running smoothly. The capsule itself then disperses heat into the surrounding waters just as a submarine does by running part of liquid cooling loops through cold water. No measurable heating of water is detected once beyond a few inches away from the outside of the data center.

In order to power the submerged vessel, a cable is run from Orkney Island that is running off completely renewable energy sources. At full operating capacity, the 864 servers can pull up to 250kW. For a data center, this is extremely low since that averages to less than 300W per server. It should be taken with a grain of salt though as Microsoft has not detailed the hardware specifications of its sunken data center.

Over the next year, Microsoft's research team will be closely watching the underwater data center to see how it is performing. Internal humidity levels, power consumption, temperatures, and other statistics will be collected routinely. Should any hardware fail during the first year, there is no way to perform a repair.

Placing submerged servers off the coasts of densely populated areas has the potential to greatly reduce experienced latency. The majority of the world's population lives near large bodies of water capable of supporting similar submerged capsules.

Going forward, Microsoft may be able to rapidly deploy new data center capsules in as little as 90 days. Installing a new data center can take upwards of two years for traditional installations, making Project Natick a game changer for providing enterprise-grade compute resources.

Permalink to story.

 
Reminds me of a similar experiment...

crap.jpg
 
It's true: it will warm the water... But if the entire unit can be cooled more efficiently, that's a power saving that might/should mitigate the direct planet warming, as compared to an on-land server mini-farm whose air conditioning sucks down a boatload more watts. Waddaya think?
 
What's the expected shelf life of these units? Salt water tends to destroy everything eventually. With the addition of heat that's only going to be accelerated.

And it's probably more confidence in redundancy than reliability.
 
Interesting concept, but data centers aren't infallible. It would seem to me the cost savings will tank (pun intended) the first time you have to yank it out of the water to do repairs.
 
What's the expected shelf life of these units? Salt water tends to destroy everything eventually. With the addition of heat that's only going to be accelerated.

And it's probably more confidence in redundancy than reliability.

Really, it only needs to last long enough to standup a more permanent datacenter in whatever city began growing fast enough to warrant a "pop-up" datacenter.
 
They will need a massive amount of redundancy. Most large hardware deployments would have dual redundancy (eg. dual PSU per server, dual NIC cards, dual network routers and hot swappable servers) where, if one part fails, the other can continue until full redundancy is restored, normally taking only a few days. SANs have greater redundancy and can usually handle the loss of a few hard drives but you would still be getting very worried if 2 or 3 drives in a single array stopped working. And it's not unusual to have a bad batch of drives with multiple failures. Even the loss of a single fan could cause a part of the servers to start overheating. Expecting all key parts to stay working for a year would be monumentally foolish so there must be a LOT of redundant hardware in there.
 
As if our ocean's weren't heating up fast enough on their own...

Remember when you were a kid and on a cold day your dad said to close the door after you came inside because you didn't want to 'heat the outside' Now you're older and you know that the heat from your house making any difference to the outside air is laughably impossible. Yet - make a similar comparison to water and adults can be fooled just like small children.

Anyway -
Cool idea MS, but where's the 'WHY' Why are they doing this? Signals travel across fiber at the speed of light. Moving your data center closer to reduce latency makes sense if you're talking about a thousand miles and you want to gain a few ms off a ping, but this sounds like a completely impracticable way to make response times barely faster.

I'm guessing this is for high-frequency traders or something...
 
What's the expected shelf life of these units? Salt water tends to destroy everything eventually. With the addition of heat that's only going to be accelerated.

I saw an episode on Dirty Jobs where Mike Rowe helped clean a 5 or 6 story hurricane barrier wall that was submerged under water (laying flat) until a hurricane was approaching and then it was lifted up to block off the water from surging into the bay. To combat erosion, they placed hundreds of removable zinc blocks (if I remember correctly) which gave off a slight electrical charge and the salt water would attack those instead of the steel structure. Every year they would take one day and raise the wall and clean the structure and replace the zinc blocks that were less than 50% remaining. The zinc blocks would normally last 5 years before needing to be replaced and the wall would last over 80 years.

I would think something similar would apply to this server and the structure should far outlast it's usefulness as they'll move onto new designs and methods in the future..
 
All that lost potential. All the places that could use free heated water, and they are going to dump it in the ocean.
 
I saw an episode on Dirty Jobs where Mike Rowe helped clean a 5 or 6 story hurricane barrier wall that was submerged under water (laying flat) until a hurricane was approaching and then it was lifted up to block off the water from surging into the bay. To combat erosion, they placed hundreds of removable zinc blocks (if I remember correctly) which gave off a slight electrical charge and the salt water would attack those instead of the steel structure. Every year they would take one day and raise the wall and clean the structure and replace the zinc blocks that were less than 50% remaining. The zinc blocks would normally last 5 years before needing to be replaced and the wall would last over 80 years.

I would think something similar would apply to this server and the structure should far outlast it's usefulness as they'll move onto new designs and methods in the future..

Very good thought, and this is the same concept that boat motors use - "Sacrificial anodes", or "Zincs".
 
Depending on how far from land this thing is positioned it might be considered international waters. All kinds of non-compliance shenanigans might be possible.
 
Depending on how far from land this thing is positioned it might be considered international waters. All kinds of non-compliance shenanigans might be possible.
I was thinking the same thing. Even in territorial waters there are questions about property ownership and jurisdiction. Does the neighboring cities and counties have any say on where these are allowed to be placed, or to tax them?
 
So you think its made of zinc blocks? LOL
I would bet that the servers do not use Hard drives as with all the bumping and bashing they would be the first to fail. I presume this stays down for approx 5 years until it needs to be completely replaced internally ? This could be a precursor to firing one up to the moon , or space or Antarctica ... why not dunk it into a river just downstream of a hydro electric dam, 2 for 1 there ...?

Or is this another "glomar" to listen out for russian/chinese subs ....
 
As if our ocean's weren't heating up fast enough on their own...

Remember when you were a kid and on a cold day your dad said to close the door after you came inside because you didn't want to 'heat the outside' Now you're older and you know that the heat from your house making any difference to the outside air is laughably impossible. Yet - make a similar comparison to water and adults can be fooled just like small children.

Anyway -
Cool idea MS, but where's the 'WHY' Why are they doing this? Signals travel across fiber at the speed of light. Moving your data center closer to reduce latency makes sense if you're talking about a thousand miles and you want to gain a few ms off a ping, but this sounds like a completely impracticable way to make response times barely faster.

I'm guessing this is for high-frequency traders or something...

Try running traceroute's to distant sites vs close sites. A thousand mile straight line could involve dozens of time-consuming (in network terms) hops between nodes (never mind re-transmits of lost packets).
 
What's the expected shelf life of these units? Salt water tends to destroy everything eventually. With the addition of heat that's only going to be accelerated.

And it's probably more confidence in redundancy than reliability.
Yes indeedy. Sodium chloride is the most destructive of the common chloride salts. Even exposure to a salty atmosphere will strip the chrome plating from a car's bumpers. (Of course, many of you here won't be old enough to remember chrome bumpers).

I was walking through Walmart, and no doubt by way of China, I came across this "fine product", galvanized salt and pepper shakers:

91I26ZVmJbL._SL1500_.jpg

These are allegedly from, "Better Homes and Gardens". God only knows what was in the Chinese pot metal which "galvanized" them, or how much of it leeches into the salt you put in your system, but it can't be good for you.
 
Remember when you were a kid and on a cold day your dad said to close the door after you came inside because you didn't want to 'heat the outside' Now you're older and you know that the heat from your house making any difference to the outside air is laughably impossible. Yet - make a similar comparison to water and adults can be fooled just like small children....[ ].....
This is really the most ill thought out statement I've ever seen you post. Of course your house heat will "heat the outside', but as an individual instance, the results wouldn't be measurable.

OTOH, haven't you ever heard about the massive warming effect on climate our, "urban heat islands" produce?

The way that works is, everyone has to pull together as a team, and open all our doors at once! In fact, with the very low R-factor of older buildings, you don't even have to open the door .

One data center isn't going to have a measurable effect by itself on the overall temperature of the ocean(s), but many might, especially if you start concentrating them accessible to large urban areas.
 
Last edited:
This is really the most ill thought out statement I've ever seen you post. Of course your house heat will "heat the outside', but as an individual instance, the results wouldn't be measurable.

OTOH, haven't you ever heard the massive warming effect on climate our, "urban heat islands" produce?

.

Well... I believe 1 house could heat the atmosphere about as well as 1 server pod could heat the ocean. Meaning, it's a minuscule and immeasurable amount. So if we're looking for reasons why this is a bad idea - I don't think we should consider it's effect on climate change one of them.

I'm sure an entire city could cause some disruption - but heat radiates into space - climate change focuses on greenhouse gases - cow farts and CO2 etc.
 
Well... I believe 1 house could heat the atmosphere about as well as 1 server pod could heat the ocean. Meaning, it's a minuscule and immeasurable amount. So if we're looking for reasons why this is a bad idea - I don't think we should consider it's effect on climate change one of them.
OK, we're mostly good so far. But, you certainly don't have to, "cool 1 house", by pumping thousands of gallons of water through a massive heat exchanger either. I don't have numbers, but I can say with reasonable certainty, cooling one server pod would equal many, many, many, window air conditioners. Not to mention home air conditioning only moves heat from one place to another. The heat created by the units themselves, is pretty much limited to whatever percentage of energy is lost during the process. IE: ceiling fans don't "cool" a damned thing, they create heat. The benefits are reaped through the evaporative cooling they effect on human skin. The humans by the way, are making plenty of heat in and of themselves.

I'm sure an entire city could cause some disruption - but heat radiates into space - climate change focuses on greenhouse gases - cow farts and CO2 etc.
No, aerosol hairspray is worse for the environment than cow farts. And cow farts only become an issue, when you need to raise hundreds of millions of them, to feed the already far, far, far, too many humans here on planet.

And no, the minuscule numbers of Teslas sold this far, won't do Jack squat in the way of environmental improvement, and certainly less than nothing toward the of salvation of our species.

You've completely missed the concept of "passive solar heating". Concrete retains heat, while trees cool the atmosphere by way of that, "evaporative cooling", I spoke of earlier.

Not to mention, "passive solar heating", was investigated as a design aspect of homes starting from perhaps the 70's. In this paradigm, southerly facing greenhouses, possibly utilizing massive clear columns containing water, would trap heat during the day, and expell it at night. Ostensibly, the deciduous trees you were forced to plant on the south side of the house as well, would shield the home from excess heat build up during the summer months.

Accordingly, heat retention by buildings in a vast urban setting, has every bit the impact of ICE powered transportation. (As they say on Wiki, "citation needed").

One thing humans have failed to accomplish, is adapting to their environment, but rather modifying.it to suit themselves. Other species aren't so fortunate. While natural selection and adaptation have created species which can live in climates humans couldn't, it also limits those animals to surviving in the same environment to which they've adapted.

Thus, one server pod could have a localized effect, driving sea animals from its immediate location, many could be detrimental to sea life across a larger area.

And before you dismiss me as a "seaweed hugger", take a moment to consider the massive damage we've already done to dry land, and the fact M$ is sinking these things, in an attempt to save more money to put in its own pockets.
 
Last edited:
Back