Hacker News new | past | comments | ask | show | jobs | submit login
Microsoft's underwater data centre resurfaces after two years (bbc.co.uk)
775 points by Qworg 5 days ago | hide | past | favorite | 372 comments





Interestingly in the original article it said that the datacenter was meant to stay under water for five years. I wonder why they have pulled it out ahead of time.

(source: https://www.bbc.com/news/technology-44368813 -- see video)

edit: just to be clear, my questioning isn't meant to be read in a denigratory way. just wondering. Also thank to Kydlaw for pointing out that it actually said "up to five years".


[1] mentions "up to 5 five years". But your question still hold

[1] https://natick.research.microsoft.com/


I think their phase 2 is going to be the five year project. This one was supposed to be a POC.

Based off https://natick.research.microsoft.com/


Presumably they got the data they needed? Whether it's worth doing at scale.

> was meant to stay under water for five years

Teddy, Vicky, Paulette, Sally and René?. Better playing safe than sorry, I suppose


Some executive probably said "I think it would be good to take stock of the experiment now to reduce time-risk and arrive at a decision point regarding future strategy".

It increasingly seems that public businesses seem incapable of making long-term plans, even rather trivial ones, if the cost of ending the experiment results in immediate gains. The driving force is the single most important metric: this quarter's upcoming earnings report. Businesses that tend to place bets on multi-year investments are either privately held (e.g. SpaceX) or are participating in specific public/private partnerships.

While I do think we are seeing the limits of what a publicly-traded company can actually realize, there are of course counterexamples: Google always seems to have multiple irons in the fire, and there are many other examples in the comments on this discussion of corporate research labs: https://news.ycombinator.com/item?id=24200764


> the cost of ending the experiment results in immediate gains.

This is oft-parroted without any hard evidence though. If it were true even broadly, most companies would be doing absolutely no R&D because all R&D is just cost in the near term.


Clearly companies can take on multi year projects like making a movie and even take on risks when doing so. They also do a lot of things called R&D with similar expectations of direct profits.

However, it’s not clear that Microsoft’s next version of windows for example is actual research rather than the software equivalent of making a movie. As such I think what people are talking about fundamental research not the kind of R&D which happens to qualify for a tax break but is mostly just the cost of doing business.

IMO, the line for what still qualifies for research is basically the DARPA self driving car challenge. Before the event it looked like basic research, afterward it looked like an engendering challenge to get there first. In 2004 nobody finished though several got close in 2005, in 2005 five teams finished and the race for commercial success was on.


You get diminishing returns holding it down there for 5 years for no reason.

If not much has happened after 2 years we're good to go. We need to get into the future as fast as possible.


The difference between 2 years and 5 is quite significant for server runtime above water. This argument of diminishing returns would hold better for placing it 1 year rather than 2, 9 months rather than 12, etc.

If the concern is that you'd see significant failure after 5 (which is likely) which could undercut your future plans, I could understand cutting the plan short.


It seems reasonable to apply Bayes theorem. If the failure rate underwater after 2 years is less than normal..

From my anecdotal experience working in data centers early in my career, you really don't really know if your failure rate for a specific type of drive/server is atypical for quite some time. This information is critical for many businesses, as choosing which drive or which architecture model can have profound cost implications.

It's quite possible that failures at the 5 year mark wont maintain a proportionate ratio with failures at the 2 year mark when compared to control. Maybe drive "A" tends to have an unacceptably high failure rate only after "X" hours of up-time.


probably didn't want to be trumped by google's zero carbon news..?

"The team is speculating that the greater reliability may be connected to the fact that there were no humans on board, and that nitrogen rather than oxygen was pumped into the capsule."

That does sound plausible. But I do wonder how much might have been due to extra care. If I were the sysadmin on the project, I probably would have spent extra time on component selection, cable seating, burn-in testing, etc. Lots of pressure for it to do well.

Edit: Unrelated, but this picture is funny to me. I don't think there's enough room to slide that server out, so I'm not sure what he's doing. https://ichef.bbci.co.uk/news/800/cpsprodpb/48D6/production/...


This:

"there were no humans on board"

Made me think of this:

"The factory of the future will have only two employees, a man and a dog. The man will be there to feed the dog. The dog will be there to keep the man from touching the equipment."


Haha I hadn’t heard that one before, that’s great.

Is there more to that quote that makes it make sense? Otherwise both the human and the dog are redundant as a pair as their purposes are circular and don't benefit the factory.

I've heard it before as "The dog is there to make sure no one touches the equipment, and the man is there to feed the dog", may make a bit more sense.

That's what makes it a joke.

I heard it originally in reference to planes getting automated. The idea is that nobody would be comfortable riding a pilotless plane, so the pilot is hired on to assuage people's fears, but assigned a dog to keep him from messing with the automation.

I assume the same dynamic is at play, but I find the entire thing to be riddled with hubris, and projective of industry for industry sake.


beep boop does not compute what is "humor"?

Remind me that i never play D&D with you ;)

You must be great at parties

Every day, a user on HN strays further into social oblivion

A friend of mine wrote a paper on the effects of cosmic rays on memory errors in data centers a few years ago[1]. He also posited they had effects on hard drive reliability.

The first thing I thought of when reading this article is whether the increased shielding from the water reduced the impact of cosmic rays on the hardware.

[1]: https://www.cs.toronto.edu/~bianca/papers/ASPLOS2012.pdf


Sun Microsystems had some Sparc servers where the on die cache didn't have ECC. And, cosmic rays created issues: https://docs.oracle.com/cd/E19095-01/sf4810.srvr/816-5053-10...

In a more dramatic example, Virginia Tech strung together over a thousand PowerMac G5s in a cluster they called the “Big Mac”, and it was one of the fastest (and cheapest) supercomputers of the world at the time. Unfortunately its lack of ECC RAM made its failure rate so high it would have trouble even booting. They dismantled it and sold the machines away one at a time and replaced the entire thing with an Xserve array instead, which had ECC.

I've always wondered whether the orientation of the memory chips in your computer made a difference for cosmic ray errors. That is to say, if their face is to the sky vs. their edge, will it affect the error rate?

You can try the experiment at home. The equipment is ~$100 : https://hackaday.com/2017/11/27/make-a-cheap-muon-detector-u...

It's also a really good random number generator, for those Hackers/crypto people that would like a true source of randomness.


Cosmic rays can come from any angle, and even pass through the entire earth without impacting anything. (millions of them are zipping through you and your RAM, right now! electrons barely dodging out of the way in time.)

I think (but have no reference) that the amount of cosmic rays the planet blocks by being in the way is dwarfed by the effects of the magnetosphere and solar rays.


Slight nit-pick: Neutrinos are the ones that are zipping through you and the Earth. They interact with things very rarely and have no real charge anyway. The rays you'd worry about with computer hardware are photons (gamma rays) with high energy enough to create new charged particles. Also other charged particles like muons and their decay cascades.

Watching an alcohool fog chamber is such a cool experience.

Not sure how you would do the math there. Edge up would maybe be a smaller target, but hits would pass through more silicon.

I think there is something to this I used to have a radio that would pickup this type of interference and days when it was heavy I would have more than usual amount hardware failures even powersupplies.

> there were no humans on board

Back when I worked at a supercomputing center, we had "operators" on duty, who were supposed to visit the machine room every 2-3h or so and check several things.

It turned out that they were the major cause of hangs and reboots of our SunSITE server (a large FTP archive) — walking on the lifted datacenter floor caused vibrations which were enough to disturb the (terrible) external SCSI connectors to multiple drive arrays.

So, I can certainly believe that statement.


They should have probably prepared two tubes like that, and flipped a coin to randomly dunk one in water, and leave the other one up?

The outcome you're looking for is pumping one full of oxygen and one full of nitrogen - at least to verify care vs nitrogen.

Yes. Ideally you'd have lots of tubes with varying all kinds of conditions.

> so I'm not sure what he's doing.

Maybe securing the vertical bars after the entire thing slides horizontally into the tube?

I'm assuming the tube has some rails they use to slide the racks in but then they have to be properly secured in place.


Nobody's mentioning tin whiskers as a failure mode. Maybe the anaerobic atmosphere slows their growth?

https://en.wikipedia.org/wiki/Whisker_%28metallurgy%29


even treat wiki mentions conformal coatings can stop it. it's it long solved at manufacture?

Our computer room had emergency breathing masks on the wall , asked why it was there I was informed about the halon gas fire suppression system.

Needless to say whenever I was there I kept having flashbacks to the first Resident Evil movie.


Microsoft patented the way to power these data centers using tidal energy in ocean itself. They patented in 2017.

http://www.whatafuture.com/microsoft-underwater-data-centers...


"All of Orkney's electricity comes from wind and solar power"

If they patented it in 2017, I'm guessing it probably didn't make it in to this one, but the assumption is that the next version may have that technology.

That's a good enough reason to pull it out, replace/upgrade the equip, and drop it back in with the new setup (new sources of energy), and if they get the whole thing to transmit wireless-ly with a 1% failures per 2 years (8 out of the 855 servers) then it's a truly set-and-forget thing. And they can be bringing them up every 2-5-10 years to replace/upgrade the HW and drop it back in.

One would imagine that the signal to noise ratio for wireless data transmission in the ocean would be pretty abysmal. That said, there's no reason why they couldn't just run a fiber optic cable out from it, as long as it's designed to rotate around said cable.

I thought of the cable solution, but that would have some wear and tear if the capsule gets to move around. I was imagining a periscope-y antenna, in the form of a buoy.

Short range high power wireless to cable. Basically cable but with a wear buffer.

Microsoft probably has enough money that they could leave the original one in, and pop down a new one with the new green energy.

(Of course, individual projects still have finite budgets.)


I assume that the project approved by the UK government specifies some requirements about environment protection and the fate of the material.

Building a new one would take more time

Yes, but they can build multiple in parallel.

It would need an antenna buoy as wireless does not transmit well in salt water.

If that was the reason, we'd know about it from the PR release. This must have been a failure.

It seems an additional hurdle is anti-fouling. Those barnacles and algae act as an insulator, making the cold-water cooling advantage less and less of an advantage.

If the surface were that hot, would sea creatures still want to live on it? I have no idea what their range of tolerance is, but I'd assume that they're well adapted for living on relatively cool rocks. And if it's not that hot, then it seems like there's not a cooling problem.

(To be clear, I'm speculating without any real knowledge of this subject, and welcome the inevitable corrections.)


The ocean is host to a very wide number of creatures and bacteria, some live in arctic cold waters, others on volcanic vents, and every range in between. There's a good saying for this, roughly translated, that nature doesn't love an empty spot — if one type of plant or bacteria won't live in an area, someone else will quickly settle in.

We have "nature abhors a vacuum" in English

I think these quotes are coming from either end of the spectrum of physics and biology and don't necessarily say the same thing.

The beauty of the phrase is that it’s universal.

Is it?

I want to talk with aliens as much as the next guy, but so far it seems like the vast majority of the universe is a vacuum--biologically speaking.


Physically too. But it's a question of perspective: glass half-empty or half-full. Are planetary systems plentiful? It seems so, as a ratio of all star systems - as far as we know at least. Definitely not, as a ratio of the total volume of space. Similarly, the idea that we are mostly vacuum, because the elementary particles are so small compared to the size of molecules, etc. But that doesn't much change our intuitions about solidity vs. emptiness in every day life.

The way I see it, space is a life-vacuum, and human nature is to abhor that, as evidenced by the long-standing goal we have of interplanetary/intergalactic settlements.

Which is to say: you’re thinking on the wrong timescale. Just wait and see. :)


No one ever said the vacuum has to be filled biologically. The universe is a vacuum but even there you get spontaneous virtual particles getting created out of nothing as far as we know.

Virtual particles aren't particles, and don't get created out of nothing. They're a mathematical trick to simplify (read: make tractable) physics in a particular formalism.

Apart from being drawn as particles on diagrams, they violate every rule of what a particle is supposed to be.

There are formalisms where they aren't necessary, for example quantum lattice models.


That's because there is not much stuff in the universe, and whatever there is, tends to cluster together under gravity. But within such a cluster, if you create empty space, something will want to rush to fill it in.

Anyone hearing one abhors a vacuum; in fact they are designed to amplify certain sounds, such as dust particles passing through the system. Consumers equate "loud" with "powerful" and this improves sales and product opinions. Yes, I'm aware this is a different kind of vacuum.

We have a new phrase, “Gravity abhors a vacuum.”

In a post-Dyson world, many vacuums rely on gravity to work.

I think that's inertia.

I am familiar with that sentiment expressed through the phrase "nature abhors a vacuum"

The irony is that this phrase predates the Torricelli experiment, in which he used a glass vial and mercury to produce a vacuum, and the eventual discovery that the vast majority of nature is a vacuum (outside of the atmosphere of planets and stars).

And if it wasn‘t for gravity, everything inside the atmosphere of planets and stars would be rushing to fill that vacuum. Nature abhors a vacuum, but that‘s only tempered by the fact that there are more powerful forces than vacuums.

That's not quite right. Even if there were no gravity or other forces, there would be no incentive for matter to go in one direction vs another, regardless of the distribution of matter around it. Over time you'd get a uniform-ish distribution, but that's a question of statistical mechanics, not one of pressure differences.

On earth, matter only rushes to fill vacuums because the surrounding air or water pressure pushes it in.


Vacuum doesn’t necessarily mean pressure-vacuum. It roughly means extreme concentration gradient, which is as you’ve described universal.

> but that's a question of statistical mechanics, not one of pressure differences.

My limited understanding was that 'pressure' is just a simplified way of speaking about statistical mechanics.


If all of the matter in the universe were evenly distributed, it would be so close to a vaccum as to make no difference.

Don't forget, even within individual atoms, the overwhelming majority of the space is a vacuum.

It also predates attempts to use dewatering pumps in mines, leading to the snarky variant “nature abhors a vacuum, but only up to 32 feet.”

The best resources for this would be the sites that talk about the marine life around nuclear power plants. I do not have links handy, but those would be good places to start.

The surface of this container won't get that hot.

>> Those barnacles and algae act as an insulator

Maybe not. Barnacles probably conduct heat at a similar rate to water. And they create a rougher surface with greater contact area to the surrounding water. And some of them actively filter water, push it around. Perhaps having a layer of barnacles woudl increase cooling.


The cooling nature of water comes from the high specific heat _and_ the fluid mixing (convection currents, pumped flow, etc).

Barnacles and other accretive life forms means that the thermal gradient drops sharply, and hence the heat flow rate drops significantly.

Take a comparison is between (say) 25C heatsink in contact with 10C water across a boundary layer that likely on the order of 10mm thick, versus what is effectively a static fluid ~50mm thick followed by the same boundary layer.

So the unmixed fluid is going to be ~6x [1] thicker, for a heat flow rate 1/6th the original. And that's assuming that barnacles do in fact conduct as well as water, something that's unlikely (water is ~0.6, human flesh ranges from ~0.2 to ~0.5)

[1] Insert a bunch of unstated assumption about the fluid flow rate, salinity, etc etc.


This is probably one of the reason they pulled it before the 5 years they are planning. The ambient temperature have increased to a level they think it will cause more server failure.

The container is huge, and the insulation by algae is not that good. Anti-fouling matters for boats because increases energy consumption when it sails. Pretty sure for the datacenter it only contributes to a few degrees C.

Is that really still an issue? From my understanding, it's sort of a solved problem. https://www.youtube.com/watch?v=-AdW030xQB4

That’s a cool video with interesting information, but I don’t think the conclusion from it that fouling is a solved issue.

You're correct, as a marine engineer, I can tell you that the biofouling problem is far from solved. To date, copper biocide works best, but is terrible for the environment. A lot of the new coatings are also 'speed-release', meaning the hull must travel a certain speed in the water before the biofouling simply falls off. Obviously for a hull that site stagnant in the water, like Natick, this type of coating won't help. Many people have tried to solve the issue passivly, and even tried mimicking shark skin (tiny chevron-shaped scales) at the microscale to minimized biological growth's ability to "latch" to the material surface. However, I haven't seen any good commercially viable progress in practice.

I'm not sure how the MS Natick's cooling system works, but if it's anything like a ship, the heat transfer surface is NOT typically the skin of the hull (although keel coolers are essentially that, but I don't see keel coolers on the Natick hull). Instead, ship cooling systems typically suck in the virtually endless supply of cold seawater, run it through a heat exchanger (often a plate type), generally with a feshwater loop on the other side that runs to hardware to be cooled. The servers would be cooled by this fresh chillwater loop.

The above system only works if a.) you have much excess cool seawater than heat infusion (i.e. you're not trying to cool a large server farm in a stationary pond or small lake), and b.) You have means to clean the biofouling that will occur on the seawater side of the heat exchanger if left unabated. The latter, as indicated by reduced cooling capacity, may be why MS needed to stop the project 3 years shy of the objective completion date.


I can't imagine a field of these (which I assume is the eventual goal), covered in biocides, would be that great for the environment.

Maybe they will release some cooling data that quantifies the effect!

Maybe the vessels could be built with a cleanout mechanism that periodically activates and brushes the outside free of algae?

You can get anti fouling that lasts for 10 years.

Maybe there’s a temp at which it’s too warm for barnacles —question is can the systems within operate at that temperature?

The system within would have to operate at a (considerably) higher temperature, since there's a temperature gradient involved and the place where the barnacles are is the coolest part of the whole structure.

Does anyone know what the environmental ramifications are of sinking presumably noisy and hot tubes into the ocean? I've been in server rooms and they aren't exactly quiet. The picture of the racks look to me like standard 1U servers.

Well, marine life is not going to give a damn about warm tubes. There's plenty of natural heat sources in the ocean way more extreme than that.

As for the noise, that can have a negative impact, but it's going to be nothing compared to the noise ships are making: https://www.npr.org/2020/07/20/891854646/whales-get-a-break-...

Whether it's high-pitched or low-pitched noise also makes a huge difference underwater.


They sunk it half a mile offshore and 117 feet deep. Hardly the abyssal zone. Other than hydrothermal vents, what natural heat sources are you thinking of?

(and yes, it's a damn shame the oceans are so noisy due to ships and sonar...and you're prob right about heat not being harmful, I'm just curious what you're referring to)


> Other than hydrothermal vents, what natural heat sources are you thinking of?

Mostly the sun, but also warm freshwater rivers (one of the biggest temperature differentials you'll find excluding hydrothermal vents). Areas with a lot of organic material (algae etc.) close to the shore can also get very warm compared to the temperature average. Though I'm not sure about the cause/effect for the latter.


Isn’t one of the environmentalist complaints against nuclear is they heat up the water near shore, which changes the ecology?

One of these might not be a big deal, but a thousand? Worth questioning.


>Isn’t one of the environmentalist complaints against nuclear is they heat up the water near shore, which changes the ecology?

This is a valid point for small streams, whose ecology can't adapt when the temperature changes mid-stream.

For oceans the picture is a bit different though.

Suppose we're looking at a cubic kilometer of seawater that has a temperature of 10C. The entire output of a nuclear power plant (single unit, about 1200MW) would heat that water by less than a single degree in a month.

The average power consumption (and thus roughly heat generated) of a single server rack is somewhere around 12KW, so you can power about 100.000 of these for that. Microsoft's submarine had only 12 racks. From this one could conclude that even the localized effects are likely to be minimal.

Now for perspective: the entire ocean has a water volume of roughly 1.35 billion cubic kilometers. Next to major energy sources like the sun, and even smaller ones like hydrothermal vents and streams of warmer water entering it, your puny server submarine is not going to be noticed. With a nuclear power plant worth of energy you're heating the whole body for just about 1/100.000.000th of a degree per year, assuming the heat wouldn't dissipate out of the water at some point.

And really, if you're running your servers outside the water, the heat would dissipate into the ocean at some point too, making it a moot point in the grander scheme of things.

So the only area of interest concerning marine life is about 5 meters in every direction from your server tube. I'm willing to bet it'll have way less of an impact than a warm freshwater river discharging into the ocean.


But isn't the real issue how these small changes add up over time? Seems kind of like saying "a single car will not put out enough CO2 to impact the atmosphere" but here we are with millions (billions?) of cars that are definitely having an impact on their environment. Granted, we will not have millions of data centers but maybe we have enough that it makes a negative impact. Certainly seems plausible.

I don't know anything about ecology, but my intuition is that all of these things from wind turbines to data centers under water, have an impact on their ecosystems. It wouldn't surprise me if we found out these had a negative (or neutral) impact on their environments.

I guess my point is that it seems naive to simply hand-wave off the possibility that these supposed environmentally friendly technologies actually negatively impact their environment. Whether or not that negative impact is less than the alternative is an interesting question.


The reason the northern line (on the tube) is so hot is due to 2decade+ of heat building up in the surrounding rock. Not the same dynamics because of convection in water, etc, but an interesting and very real example of how local buildup of temperature can be hard to manage over multiple years.

Talking, talking, all that talking, fake stop signs on any direction we could move on.

We have a freakin' enormous open thermonuclear power plant beaming on us since the beginning of days. Just don't impact the planet's ingress/egress ratio with that CO2 and it will all turn just fine!

Heat - not a problem now.

Greenhouse gas - big problem now.


I wonder if the heat and noise of even a tenth of the data centers in the world would be still so harmless.

sad,the way you have presumed marine life behaviour.

If you can dump thermal energy into the ocean I suspect there wouldn't be much need for noisy fans.

I was curious about this. The trick is moving the heat from the components (CPU/GPU/RAM/PSU/storage/etc) to the hull. It's full of nitrogen, but the heat would conduct much better via liquid (heat pipe) than gas (fans). The photos make it look like these are standard server racks, though, so there must be fans pumping the heat away from the components. Right?

Unfortunately, there are no oceanographic environmentalists with sufficient clout to have any say, whatsoever, in this unaccountable private endeavor. I doubt the Scottish Environment Protection Agency or UK's Environment Agency were invited to be veto-empowered stakeholders.


I must ask, apart from the renewable energy factor, why do it in Orkney?

There are many islands and archipelagos in Scotland, the Hebrides which are closer to the mainland but still out of the way of fishing.

Many of them are going to be setup with tidal and wind generators.

Would be interesting to see how it was tethered to the shore with networking and power.

https://www.theguardian.com/environment/2011/mar/17/10megawa...


I've heard about Orkney quite a lot, I think it must simply be an innovation hub.

That I can think off the top of my head:

- Orbital marine, making tidal turbine https://twitter.com/Orbitalmarine

- Their grid https://www.youtube.com/watch?v=FXe1hBvlylw

- Their Hydrogen facilities, using extra renewable https://www.youtube.com/watch?v=Rybpaqhg5Qg


It's a place where a lot of research around this theme is conducted.

It's apparently powered by energy from the European Marine Energy Centre:

http://www.emec.org.uk/

I suspect there is simply more infrastructure for this kind of thing in Orkney that the other islands.


Depending on the exact shape, sometimes dropping giant bits of steel in the ocean can be very good for shipping; we purposefully sink ships that are EOL for this exact purpose fairly often.

That's the first I've heard of something like this. What's the purpose of dropping specifically-shaped steel into the ocean?

Oh, that's so that the drop-shippers can increase their throughput.

IIRC, it gives coral a place to grow.


As someone else said, it helps coral reefs grow.

And I meant to say that it’s good for fishing, not shipping. My bad.


The thing is that if putting data centers underwater solves the cooling problem, all the heat produced is wasted, so it this really a good solution?

Heat is awfully hard to transfer in the first place, especially when it's not even that hot. It's handy if you can put your data center underneath a swimming pool, for example. But I'm not aware of any large-scale heat recovery projects from data centers. Data centers generally spend extra energy to remove the heat...

But honestly who cares if your power comes from renewables in the first place -- solar and wind? It doesn't seem right to even frame it as "wasting" heat in the first place, anymore than the sun's heat was being "wasted" warming up the ocean in the first place.


The new LUMI EU supercomputer in Finland will put its waste heat into a district heating system: https://www.hpcwire.com/off-the-wire/csc-lumi-supercomputer-...

I'm sure smaller scale data centre heat to district heating schemes must already be in place. Fundamentally you are using the same technology to cool the data centre (a heat pump), just pushing that heat into hot-water / steam, rather than dumping into the air.


This has been active in Stockholm for years, at least:

https://www.opendistrictheating.com/


Also Wedos here in Czech republic; https://www.wedos.org/en/

There are towns in Denmark getting heat from Google and Facebook datacenters so I don't see why it shouldn't work elsewhere. The way MS does it here the heat is 100% waste and should be taxed like other waste. Not to mention what about the noise? The sea is already noise polluted.

Here's a source (in Danish). Wind energy in -> 25 MW/h of heat out (to heat up 12.000 houses):

https://www.computerworld.dk/art/252732/facebook-vores-data-...


You want to tax unused heat? Are you serious?

If entropy continues to be an issue taxation and regulations are sensible solutions.

If the power is generated from waves, wouldn't the energy would have gone to waste at the shoreline/breaker anyways?

> I'm not aware of any large-scale heat recovery projects from data centers.

I don't know what you'd call large-scale, but the school I went to was entirely heated by the datacenter across the road and the DC still had plenty of heat to spare. Afaik they heat all of Science Park Amsterdam with the Equinix DC just to the east: https://osm.org/go/0E6VOVkQc-. I think it was a requirement of the local government that this waste product be reused. Amsterdam has also had a datacenter stop due to power supply issues.

Edit: this seems to be the press release (only source I could find): https://www.equinix.com/newsroom/press-releases/pr/122801/eq... "Third data center in amsterdam" matches that this DC is called Equinix AM3, and it says "Excess heat from the data center will be used to warm nearby buildings and for other third-party uses."

Not to say that this happens with all the data centers across the world, but if using a decent chunk of a DC's waste heat isn't a large-scale heat recovery project, then I'm not sure what would be.


Well, you could put a sterling engine to transfer the heat to the ocean and generate energy back. This is exactly what some submarines do [1]

[1] https://en.wikipedia.org/wiki/Applications_of_the_Stirling_e...


I don't know if it's the best solution, but it would seem to beat the "use A/C to cool it" solution.

The heat produced in a traditional data center is already unwanted and wasted and takes a ton of electricity and machinery to get rid of. Given that processors like to operate around 50 C, that's not a lot of heat to preserve and transport and do something useful with.

Maybe design electronics that are happy at higher temperatures?

The issue is more that transistors will always generate some heat- making the threshold higher / being more efficient doesn't solve the fact that there's still some waste thermal energy that needs to be removed.

Modern transistors are rather small and would desintegrate at too high temperatures. Not much you can do about it at those scales, any material will decompose.

It's literally waste heat, what else are you going to use it for? You won't efficiently generate electricity with it or anything like that. About the best you could hope for is warming up inhabited spaces in winter, but that ends up not being cost-effective because you'd rather have the data centers in the middle of nowhere where land is cheap (and you can't really transport the heat to where it's actually needed).

It isn't wasted it is helping to heat up the oceans.

That's the cruel part of the thermodynamics. The waste heat here is not hot enough so it's really difficult for the recovery effort to reach a reasonable efficiency economically.

That's why you don't see it...

https://en.wikipedia.org/wiki/Carnot%27s_theorem_(thermodyna...


Sure you see it. This is just a way to try to get rid of waste for free. Both Google and Facebook (Apple too I believe) delivers heated water to heat up houses from their datacenters. The one near me (Facebook) takes in wind energy and produces 25MW an hour and heats up 12.000 houses by wind energy and waste heat.

Very interesting to see a lower failure rate. Is it simply a function of possibly better thermals than your average datacenter? Less EMR?

Other explanations:

- Technicians were extra careful (slow) when installing the equipment.

- The datacenter pod used no recycled parts. Traditional datacenters are full of recycled marginal-quality parts. Maintenance teams balance the cost of buying new parts, the cost of testing used parts, and the labor & downtime costs from recycled parts failing.

Extreme reliability is already achievable but not economical. One reason why Google Search beats Bing is that Google's infrastructure software is more tolerant of flaky hardware, so Google can spend less money on hardware maintenance, reducing the cost per search.

Hopefully Microsoft will release a report and tell us the source of the underwater datacenter pod's low failure rate.


Besides reasons mentioned in the article (no oxygen, no humans to bump into stuff), I wonder if radiation could play a role. Cosmic rays, maybe just less radio sources around (tons of other servers).

The container was filed with Nitrogen, so no pesky Oxygen corroding everything.

From the article:

>The team is speculating that the greater reliability may be connected to the fact that there were no humans on board, and that nitrogen rather than oxygen was pumped into the capsule. > >"We think it has to do with this nitrogen atmosphere that reduces corrosion and is cool, and people not banging things around," Mr Cutler says.

Better cooling and isolation from all kinds of radiation seem like they'd also be beneficial.


Impossible to touch it. Better (oxygen poor) environment for circuitry.

Another example of the journalistic malpractice of not linking to the original articles or data.

https://natick.research.microsoft.com/

https://news.microsoft.com/innovation-stories/project-natick...

One of the facts left unmentioned is that this was built and operated by Naval, the French state-owned submarine and shipbuilding corporation.


> journalistic malpractice of not linking

That disease is very prevalent, and I don't understand why.

My less-cynical guess is that the industry is stuck in the past, and journalists need better training and tools.

My more-cynical guess is that they are afraid of irrelevance, so they are defensively trying to keep you in their walled garden of information instead of encouraging you to get into the habit of getting info more directly.


I've always assumed it's the fear of dead links hurting their SEO.

Outbound links also remove the reader from the newspaper's ad revenue stream.

I don't get your point? Why is it that relevant to you?

It rewards originality over the long term rather than the fastest copy and paster.

Sort of like how we try link to the original YouTube video.


Understanding what is OC is an increasingly important point for digital content makers, from memes to ideas for explainer videos.

I think with actual articles written by presumed journalists, linking to source data is what establishes the credibility of the author’s writing and suggests they have read and understand the content.

Not linking to it doesn’t mean the author doesn’t understand it, but it may mean their work does nothing more than regurgitate (adds nothing of value apart from increased distribution)


1. Linking to your sources is a good practice. 2. Their article adds little if any insight over the original article. I would often prefer to see the source and read it instead/additionally.

Breaking news: Experts on underwater vessel construction asked to construct underwater vessel.

Is it not significant that this level of expertise is required? Do you think Microsoft paid Naval, or did they want to acquire expertise? Why use Naval rather than a civilian focused fabricator?

When you consider the significance of undersea cables to the global economy, the propensity of states to intercept them covertly, the difficulty of attacking or even finding submersed compute, and so on, the ramifications are significantly greater than 'green compute'.


Especially scummy when they've also broken the functionality of the back button and all the links are to other BBC articles.

It's like the main function of the site is to trap you.


"Please don't complain about website formatting, back-button breakage, and similar annoyances. They're too common to be interesting. Exception: when the author is present. Then friendly feedback might be helpful."

https://news.ycombinator.com/newsguidelines.html


If you're in the US like I am, I think it's because the BBC redirects all non-UK visitors to BBC.com, and so when you hit the back button from BBC.com you get sent back to BBC.co.uk which in turn redirects you to BBC.com. It gets on my nerves too.

Could be but I am also seeing a trend where some websites manipulate the navigation history so that when you click the back button in your browser you end up at their home page even if you came to the page you are currently on via some other site.

I first saw this on Facebook but I have since seen even sites that I used to respect follow this same pattern.

To me this is nothing but another dark pattern.


Wouldn't it be nice to have a browser engineered to protect against that? I'm tired of struggling with dark patterns every day, from URL hijacking to modal dialogs with deliberately broken layouts to scrolljacking.

I thought back-button hijacking, malicious or accidental, had been solved in browsers years ago. I suppose I need to go search for an extension to fix that instead.

"We are hopeful that we can look at our findings and say maybe we don't need to have quite as much infrastructure focused on power and reliability."

I was left wondering whether they were referring to the project, or just Azure Availability in general. /s


Was the load in the data center synthetic, or did it handle production traffic? Curious to see how they split it out.

I doubt there's a "Orkney-Underwater" region.


The official site has the info you are looking for.

"Natick was used to perform COVID-19 research for Folding at Home and World Community Grid."

Also the data center designation for it was "Northern Isles" (SSDC-002).

https://natick.research.microsoft.com/


Almost certainly synthetic

It would be nice to see some numbers on energy used to power this versus a regular set of servers.

Also nice if they discussed the energy/cost involved with deploying and retrieving these capsules and how well that would amortize if this became a commercial solution.


From https://natick.research.microsoft.com/ Phase 1 demonstrated the feasibility of the subsea datacenter concept, including our ability to remotely operate a Lights Out datacenter* for long periods of time, operating with a highly efficient PUE (power usage effectiveness is total power divided by server power; lower values are better, 1.0 is perfect) of 1.07, and using no water at all, for a perfect WUE (water usage effectiveness is the liters consumed per megawatt of power per minute; lower values are better, 0 is best) of exactly 0 vs land datacenters which consume up to 4.8 liters of water per kilowatt-hour. For Phase 2, our goals are to:

Develop one full scale prototype subsea datacenter, which could be used as a modular building block to aggregate subsea datacenters of arbitrary size

Gain an understanding of the economics of undersea datacenter TCO (total cost of ownership is the full lifetime cost of a datacenter including manufacture, deployment, operations, and recovery) should we proceed to commercial deployment.


Why heat the oceans indirectly when you can just place the heatsource directly in them? Much less chance of losing precious ergs.

Seriously though, what is the direct ecological impact of doing this at scale, would the local increase in temperature have an immediate effect on the life around it? If so how much of an impact?

What about the effect on surface life and life in intermediary layers of the water? After all, a body this size radiating 10's of KW of heat would cause substantial convection. At data-centre scale could it conceivably shut-down ocean currents or re-route them?


An HVAC system above ground would generate far more heat. The advantage of sinking the data center would appear to be that it can function with passive cooling from the surrounding water only, no HVAC. This means less electricity consumption (meaning less energy generation required, and fewer resulting greenhouse gasses). And HVAC systems create far more waste heat than they do cooling, so this passive cooling will just be moving the source heat to the water, not dumping extra waste heat as well.

Overall an underwater data center should generate far less heat overall than an aboveground one.


I think the question is what is the local impact of the temperature increase? It would be good to know if it harms certain types of ecosystems ahead of time rather than after the fact.

It's also possible that a direct increase in ocean temperature has undesirable knock-on effects that don't take place if you operate on land.


I believe it would be less impactful than many power plants that vent their heat into the ocean. There are many studies, mainly around marine life (phytoplankton, manatees, other endangered species). These systems probably produce similar heat output into the water.

https://assets.nrdc.org/sites/default/files/power-plant-cool...

https://www.sciencedirect.com/science/article/abs/pii/S00253...

https://res.mdpi.com/d_attachment/water/water-11-02577/artic...


Well this was all an experiment itself and I am sure the ecological factor is being taken in consideration

Well they why not create artificial bodies of water and just keep our data centers there? How much water is sufficient for the cooling cycle to not require additional energy? How much water will need to be added to such pools to deal with evaporation?

I think on a smaller scale test with an artificial pool would provide some solid answers. We could even put solar panels across the surface to limit wind and solar evaporation effects


You'd have to pump loads of water for it then? to fill the pool and replace what is lost to evaporation.

"I think on a smaller scale test with an artificial pool would provide some solid answers. We could even put solar panels across the surface to limit wind and solar evaporation effects "

If it's artificial pool you'd need evaporation to cool it.


Do you not consider lake Powell to be an artificial body of water?

never heard of it until now

The issue is CO2, though, not heat per se. Yes, the two are correlated, but if you use renewables to power your things, it doesn't really matter how much heat you generate.

Please correct me if I'm wrong.


> but if you use renewables to power your things, it doesn't really matter how much heat you generate.

Thermodynamics should really be emphasized in schools. You are getting some amazing responses that are completely ignoring the fact that our renewable energy solutions are not increasing the overall heat in our planet. They are just moving energy around.

The only thing that really matters is how are the emissions going to look like when we are manufacturing renewable energy equipment (either new capacity or replacing faulty ones).


That's faulty. Solar installations change the albedo of the planet. That's not a problem if the fraction is a very small one and it is spread out. But if it is concentrated or relatively large it certainly could have an effect.

> Solar installations change the albedo of the planet

They do. And so does every single thing we build that's exposed to the outside - bulding, cars, even you when you are out and about.

The effect is minuscule unless we are turning the planet into Coruscant or this is a gray goo scenario.


>The effect is miniscule //

The effect of a human on the planet is miniscule, but we're still in the bad state we are.

We're looking at covering the planet in solar panels, worth considering how the albedo changes will effect things on a planetary scale.


We thought much the same when we introduced the automobile. Plastic for packaging purposes. Freon to help with the Carnot cycle in refrigerators. These things don't matter when you do them once or twice. But when you start doing them on an industrial scale it changes the equation from 'no effect' to 'unknown effect'. And unknown effect might be anything from negligible to planet wide catastrophe. It would be nice to know where we land before taking off.

Oceans warming up is a big thing, and local effects can be substantial even if global average change is negligible.


Isn't increasing the albedo cooling the planet down?

Yes, an increase in reflectivity would cool the planet down. But solar panels actually absorb a lot more than they reflect (they would have to). The ideal solar panel would be utterly black.

Ah, yes, that makes sense. The reflective cover they have made me think they reflected more than they absorbed (which, yes, doesn't make sense).

You're right that renewably-generated waste heat isn't a big deal all-else-being-equal; but all else usually isn't equal, and climate change is a always question of numbers (my favourite example: extracting a barrel of oil from the air/flue using renewable energy and sequestering it long-term, compared to the cost of leaving a barrel's worth of oil where it is in the ground; the latter is free, except for opportunity cost)

Hmm, wait, why isn't the first one free in terms of CO2?

The first one requires development, construction, installation and maintenance of equipment; it requires explicit effort from many people; it will need to be decommissioned at some point; the storage will have some marginal cost (e.g. containers, drilling holes, etc. depending on the method). These things all cost CO2, unless the economy has already been decarbonised (in which case it wouldn't be needed).

More importantly, being "free in terms of CO2" is still an all-else-being-equal perspective. It's focusing on one aspect (CO2 emissions) of one small cog (a CO2 extraction+storage plant). If we look more broadly, each barrel extracted is offsetting less than one barrel being burned elsewhere (since nothing is 100% efficient). CO2 extraction and sequestration is thus a form of power transmission: the work that is required to offset emissions (e.g. from a car) is being performed away from where the emissions are made (although for flue capture this might be quite close!). For example, we can think of these as being roughly equivalent:

- A fossil fuel car with solar-powered carbon capture and storage onboard

- A fossil fuel car with solar-powered carbon capture and storage in some other location

- A solar-charged battery-electric car (+ a little CCS to offset manufacturing emmissions, etc.)

These are all solar powered and carbon-neutral (as long as they offset enough). Let's say they each receive a similar amount of solar energy: the first will not get very far, since offsetting is very energy intensive and it needs more fuel to carry the solar+CCS equipment. The second is more efficient, since the fuel doesn't need to move the solar+CCS equipment; it's as if the offboard CCS is transmitting a little extra power to the car. The third will get much further, since the battery and electric motor make much better use of the solar power than the CCS system.

The first approach is clearly silly. The second is useful in situations where renewables can't be used directly (e.g. jumbo jet fuel), but is incredibly wasteful and expensive compared to the third. The third approach is best, and should be used as much as possible.

If somewhere has an abundance of renewable power (e.g. geothermal in Iceland), then "transmitting" it elsewhere via CCS is much less efficient than, say, laying a high-voltage DC line; or moving high-energy, location-agnostic activities to the region like aluminium smelting or datacenters.


> it doesn't really matter how much heat you generate

Doesn’t matter is strong. It won’t in the short term. But as we continue increasing our energy use as a species, the simple thermal problem of waste-heat management will certainly surface.


If you use renewables, you're using heat that is already around on the planet. As long as you don't change planetary albedo, equilibrium temperature is the same.

Of course, heating water locally, etc, can cause its own environmental impacts.


This depends on the renewable.

Hydro, wind and waves are probably at that ideal except to the extent that they are tidal energy.

Solar panels... are literally in the business of making the planetary albedo higher, to the extent that they do so they are introducing thermal energy.

Geothermal is in the business of increasing the rate at which heat escapes from underneath the surface, which increases surface temperature.

Tidal energy is in the business of extracting energy from the kinetic energy of the moon, which probably increases the temperature of earth (but it's hard to say to what degree).

Fusion (if it ever becomes practical, and you count is as renewable) is in the business of releasing potential energy trapped in hydrogen atoms, increasing the temperature. This is particularly problematic because fusion would also enable us to increase our energy usage to the point that direct heating becomes a problem at the same scale as CO2 release currently is.

Fission (if you count it) is like fusion.

Space based solar (if it ever becomes practical), is increasing the area of the sun captured instead of the albedo, and is directly introducing energy.

Etc


this! another way to look at it: your solar panels are temporarily "stealing" heat generated by sun rays hitting a surface of equivalent color. When you use electricity to do some work, it will be turned back to heat. so if you want to be 100% "heat neutral", all you have to do is to ensure that for every square meter of solar panels you also paint an proportionate area with a color that reflects the right amount of sunlight to compensate the difference between the color of the solar panel and the area that was there before you installed the solar panel.

If you think that's silly (and rightfully so), then perhaps that can sharpen your intuition on how insignificant is the total amount of heat produced by our devices (even if cumulatively they are a big looking number); the total amount of radiation that comes from the sun down to earth is staggering.


Solar panels don't work on IR, they work on visible light.

IR might be reflected or absorbed.

The visible component is absorbed. Whether it would have turned into heat or reflected back in to space is about a 1 in two chance.


sorry I don't understand the comment; visible light carries energy too and thus can heat materials it hits. what happens to the IR part of the spectrum is irrelevant and can be captured by the general concept of "color of the object", defined as the spectrum of reflected vs absorbed EM radiation

"Renewable" refers to energy, not just heat. The sun's energy is be used for other purposes than heat, thankfully.

Growing a 100 trees and chopping them into lumber is less hot than growing 100 trees and burning them.

If all the sun's energy were converted to heat (and not radiated away), we'd be in big trouble. That's what "carbon" pollution is all about -- Carbon dioxide is a greenhouse gas that traps heat. Reducing albedo is one way to increase tempterature, but directly burning stuff is another way.


I think you missed the point of what I said. When we're talking about powering data centers with renewables, talking about sequestering carbon via lumber is rather orthogonal.

The point was, coarsely: using e.g. solar panels only changes the Earth's surface temperature to the extent it changes albedo. (Ignoring second-order effects of concentrating heat and associated effects on radiation, etc.)


Not necessarily - if (I think this is true but am not able to prove it) the earth is on a slight negative carbon slope, and renewables reduce that slope we are still making an impact on long term temperature.

Actually, in the long-term, waste-heat management is the only problem. Every other problem can be geo-engineered away, but we could never geo-engineer away thermodynamics.

Sure, where "doesn't matter" read "is way way less important than CO2 emissions".

When you say "short term," how long are you talking about?

The similarly extreme version is kind of an interesting comparison, and similarly irrelevant.

The 84,000 ppm for 60 minutes is roughly the lethal CO2 concentration. Local CO2 concentration is often several times atmospheric CO2 levels. That’s clearly addressable but I suspect around 8,000 ppm atmospheric we would start to see deaths from this which is achievable from coal deposits. Reaching a fully lethal atmosphere is of course much harder.

So, I think you’re right temperature pollution at extreme levels is worse.


Also consider cost (in terms of CO2, resource consumption, etc) of building and maintaining the renewable energy sources.

I don't know how this would compare between underwater datacenter, solar panels, wind.


Building and maintaining the non-renewable energy sources also costs CO2, though.

> And HVAC systems create far more waste heat than they do cooling, so this passive cooling will just be moving the source heat to the water, not dumping extra waste heat as well.

Where do you get that idea? A typical EER 12 air conditioner will move 3.5x the heat energy that it consumes (COP = 3.5) [0].

[0] https://www.engineeringtoolbox.com/cop-eer-d_409.html


Another way to put this is it puts 130% of the source heat out into the environment.

No, it's way worse than that.

At the location of the hvac it puts 130% of the source heat into the environment.

But 30% of that heat put into the environment came from electricity generated in a power plant. Power plants are typically less than 50% efficient, so it put's out as much heat into the environment at the source of the electricity. Bumping the value to 160% (130% + 30%).

However waste heat is a small fraction of the heating that electricity generation produces. Very roughly 10 times as much heat is trapped via the CO2 released than heat is released by the power plant. Bumping that value up to 460% (160% + 30% * 10).

I.e. 4.6 units of heat are put into the environment for every unit of heat removed from a closed system.

(Obviously the details of this depend dramatically on the environment. Heat pump efficiency depends on the degree of temperature gradient, CO2 release and power plant efficiency depends dramatically on where the power is coming from, which changes with where you are located.)


Fair points. IMO, the 460% metric is a bit of a wacky number, though, because of not counting the carbon and efficiency involved in the original 100%. Assuming the 100% comes from the same power source, it's still only 25-30% "worse".

Also I think you're a bit pessimistic about modern power plant efficiency-- combined cycle plants do better than 50%, and that's before we're considering any benefit from renewables.


> Assuming the 100% comes from the same power source, it's still only 25-30% "worse".

Fair point, I guess my argument makes more sense if we were discussing moving naturally occurring heat out (i.e. household ac) than with respect to cooling a datacenter.

Nitpicking the numbers used in the estimate... is probably not worth it. Every bit of it is a very rough order of magnitude number. If you're somewhere with 95% renewable energy it should be an order of magnitude better, if you're somewhere where energy production is dominated by an inefficient coal plant it should be an order of magnitude worse.


Isnt this assuming coal / gas powerplants ?

Wouldnt solar / Wind have a smaller CO2 foot print and hydro electric be more efficient ?


I don't see the contradiction. The heat it moves plus the heat generated through inefficiency. It's always > heat moved.

I guess it depends what you mean by "waste heat". I consider the waste heat as the "extra" heat/energy on top of the heat moved. But even if you consider all the heat as waste heat, calling 30% more "far" more seems like an exaggeration.

Depends on the reference. If the underwater datacenter produces only e.g. 3% atop, 30% is still "far more".

But without scientific sources, these are only wild speculations.


Really, you have to isolate the variable you care about here, which is 'how much energy is spent on cooling.'

Consider, if you can achieve a fully passive cooling solution by dropping a datacenter into a lake, you've reduced the energy consumption in service of cooling by 100%.

(In reality, water cooling isn't "free," but I'm willing to bet the amount of energy required to dump heat into surrounding water is a whole lot less than the amount of energy spent for the compression cycles and forced air of above-ground HVAC systems. Water cooling using direct application of chilled water is already a thing, using lakes or retention ponds as places to dump heat; what being at the bottom of a lake gives you is a more consistent and proximate source of cool water than you might expect from a current chilled water distribution system)


It's not just the HVAC system itself, it is the supply chain to operate the HVAC as well. There is waste heat from the power transmission system to get power to the HVAC unit, waste heat from the power plant that produced the power, waste heat from the transport of raw materials to the power plant, increased global warming heat from the CO2 that the power plant produces while burning those raw materials, etc.

Obviously this can be mitigated if you are able to get renewable power from a nearby source like a geothermal plant, hydroelectric, or solar. But if you are using fossil fuel power from a long distance away, that means any unit of heat moved by an HVAC involved many units of heat production to ultimately move that unit of heat.


For every 3.5 BTU moved, it consumes 1 BTU. That BTU goes somewhere. Heat in the inductor coils soaks into the environment, etc.

saying ‘at scale’ is being unclear, because it seems to be implying that we should consider what would happen if we put every data center today underwater in the same spot, which is what no one is thinking of doing...

You could have a solar powered HVAC system that captures the condensed water, and then submerge the thing you’re trying to cool in that water!

P.S. am not HVAC wizard.


The issue is the HVAC system itself creates a lot of heat. It normally doesn't matter in a traditional system as the heat is generated out side the areas being cooled.

I'm sorry but this is such a 'goalpost' movement here that simply takes away a lot of context. The whole point of doing a dc underground is to reduce the hvac energy consumption which has a much larger ecological impact in terms of heat generated, power usage, potential greenhouse gas usage.

The name of the game here is energy efficiency and conservation. Use less power by reducing power distribution loss at scale. Want an even greener solution? Make em nuclear powered like a giant submarine- that way the power generation isn't creating a heavg ecological impact.

I think your posing of these questions genuinely fail to appreciate this for what it is, a successful proof of concept that will permit a step toward a greener future.


I don't think jacquesm is off the mark in asking these questions, especially:

> what is the direct ecological impact of doing this at scale

What is the impact on sea creatures if we put these giant heated cylinders in their territory? Is there any impact? I really doubt we have the answers yet.

> at scale

...also implies that there will be more than one. So if one heat cylinder doesn't do anything, what about 10? 100? 1000? 10,000? The cylinder pictured is really quite small, you'd need to make a ton of them or drop one mega-cylinder to compete with land based centers. And it's not like our current data centers are keeping up with computing demand; we are building more today and presumably will continue building them decades into the future.

So these are extremely important questions. We shouldn't dismiss this as a workable idea but we should also keep in mind that fucking up the ocean is a far more consequential action than fucking up some acres of land.


There are hot thermal vents and underwater volcanoes throughout the oceans. Creatures adapt. The oceans are huge. Yes, if they place them in ecologically fragile environments that would be a problem —the solution is put them offshore where they don’t bother anything.

> Creatures adapt.

You're missing half the saying. Creatures adapt, or die. Given the recent articles about how species extermination is accelerating, it's a valid assumption that as many are dying as are adapting.

Not to mention, oceanic currents are propelled by heat. Changing that balance (whether by man-made resources or natural sources) could change the currents, which would impact coastal weather patterns.

https://e360.yale.edu/features/will_climate_change_jam_the_g...


Far more of earth is covered by oceans than by land. Why are these questions more relevant when building an underwater DC than a ground-level one?

They're not more relevant, but they are equally relevant.

This is adding the ocean as a new layer in the overall heat dissipation stack for datacenter computing; it's impacting a whole new ecosystem that it wasn't before. Thus, we need to be asking these kinds of questions about how this will impact oceanic ecosystems.


It is adding... the question is how much compared to natural systems that add heat to the oceans already (in addition to sunlight, thermal vents, oceanic fissures, underwater volcanoes, etc.) is it significant or is it insignificant?

> is it significant or is it insignificant

Then a study (ideally multiple studies) should be commissioned to find this out. We shouldn't just do it because we can, and because armchair physicists are pretty sure that it's insignificant.

Because, that's what Microsoft (and their partner) did. They did it because they could.

Remember when we, as a global society, believed that CFCs would have an insignificant impact on the environment? I do. Perhaps we should tread a bit more lightly when we're already running into species extinction and global climate change issues.


I doubt these would add more heat than a single new volcano does, but I’m all ears. According to research[1] tectonic activity adds about 3 cubic km of magma/lava each year to the oceans (that’s 100 times the annual production by mauna loa.)

[1] http://volcano.oregonstate.edu/book/export/html/138


This sounds a whole lot like the "global warming couldn't possibly be our fault (because: volcanoes) so let's not change anything" line of thinking, just applied to oceans instead of the atmosphere.

In which case, we'll just end up talking past each other, so I'll wish you a good day.


This is only heat. Unless it’s releasing unbeknownst to me greenhouse liquids or gasses, it’s not the same argument.

Sounds like a good ~2050 startup, a company that plugs volcanoes to sell green credits for underwater server farms ;)

Unfortunately for their scheme magma would find a weakness in the crust somewhere...

They're not? People have asked these questions for decades, hence why they're exploring new options....

It's not a poor question, but the phrasing implies that the ones constructing this test datacenter have not thought about this. Given that the engineers in question have spent many months designing and constructing this device, it's extremely unlikely that some random commentor on HN (even one with interests as varied as Jacques) is going to have questions that have not come up yet.

Asking the questions here on HN instead of in a place where the designers are likely to read them makes it seem it's more of a karma grab than a reasonable "I have concerns" type situation that he really wants to do something about. Nor are random HN commenters very likely to have big, fact-backed contributions btw. It's just alarmism under the pretense of innocently asking questions.


I don't go around chasing the scientists that have put this together because they likely have better things to do with their time.

I have no assumptions either way, and if they do happen to frequent HN the chances are actually better to get such questions answered here than anywhere else.

Karma grab? FWIW I offered to hand back in all my karma points because they are utterly meaningless but Dang wouldn't have it so please spare me the nonsense accusations.

HN was still free to write to, contribute and ask questions of last I checked, I don't need you - nor anybody else - to tell me what I can or can not do here, nor do I need you to try to put me in a negative light for trying to understand something better.

FWIW humanity has an extremely well developed skill called problem solving. We can do just about anything in the laboratory. But when scaling up those laboratory experiments we often find out that what we thought was a neg positive ends up being a net negative. Before we sink a few 10's of thousands of data centers onto the continental shelf I'd like to know the ecological impact, even if that has already been studied (which I'm actually not aware of).

See also: plastic, freon, lead (in gasoline) and a whole raft of other things that seemed like a great idea at the time but for which we did not have the long term predictions when they mattered most: at the beginning, mostly because people did not ask the right questions.

Scientists in the beginning of the previous century: "Plastics, they last for ever! yay!" and a hundred years later "Plastics, they last for ever! Oops!".

Anyway, the steady stream of quality answers in this thread proves you more wrong than I ever will but this comment reflects poorly on HN, me, and ultimately, on you.


The scale is off. Things that affect local ecosystems are cooling multi-gigawatt nuclear reactors with nearby ocean water, you would be lucky if one of these drew a megawatt of power.

That said, the physics are "heat is heat". If you put it into the air or you put it directly into the ocean the only way it leaves the planet is by black body radiation. As a result locally heating some seawater nearby has (on a global scale) the same impact as heating the air the same amount.

Now we know there are some ecosystems in the ocean that prefer thermal vents and you might find that around the data center itself you have a wider variety of sea life than is found in the general vicinity due to different thermal conditions. Not sure if you could map out that was a positive or negative change.

Generally though, the ecological impact of doing this at scale is not going to be different in scale than land data centers.


Sure but one thought experiment is what if we put more of our datacenters underwater?

I expect AWS us-east-1 now adds up to over a gigawatt of critical load so if submerged in the ocean off Virginia what would be the effects, and how would those be different from terrestrial datacenters?


From a total energy gain/loss it doesn't change anything. Heat generated on planet diffuses into the ocean (the largest heat sink) and radiates into space at a (relatively[1]) fixed rate.

Given the mechanics, the overall ratio of heat in the atmosphere vs heat in the oceans is fixed by the Rtheta of the atmosphere/ocean boundary. If the atmosphere gets warmer, more heat is transferred. If the atmosphere is colder less heat is transferred.

The other question (which Jacques alluded too) is what about the local conditions. And here to the thermal mechanics give our underwater data center an advantage. Given the thermal conductivity of water, and seawater in particular, heat dumped into any spot rapidly diffuses to the rest of the ocean. That is not the case with air, which has a much lower thermal conductivity. Dumping lots of heat into unconstrained air locally can cause a localized "hot spot" which creates an interesting thermal plume and localized winds as cooler air around it comes rushing in.

To get a good understanding of just how effective the ocean is at diffusing heat, consider any of the hydrogen bomb tests in the Pacific. Prodigious amounts of heat dumped into the ocean creating a local hot spot (and a lot of steam!) and an undetectable change in overall ocean temperature. Kilauea volcano, same effect.

The ocean has a lot of thermal mass, without something like an asteroid from space, its hard to move the needle on its temperature overall.

[1] There are some pretty interesting thoughts around using harmonic resonators to convert ambient heat into IR radiation at a wavelength that can more easily pass through the atmosphere but those are just lab experiments at the moment AFAIK.


> The other question (which Jacques alluded too) is what about the local conditions. And here to the thermal mechanics give our underwater data center an advantage. Given the thermal conductivity of water, and seawater in particular, heat dumped into any spot rapidly diffuses to the rest of the ocean.

Thank you, it is precisely that which I was wondering about.


> the only way it leaves the planet is by black body radiation

But, if you're heating the oceans, you're putting an entire other mass (the atmosphere) between yourself and space. So you're heating the water, the water is heating the atmosphere, and the atmosphere is radiating that into space. It's adding a step to the overall cooling process.


OTOH, if these are passively cooled, you are not dumping the waste heat of the HVAC systems, as well as the power user to keep them running.

That's not really how it works. The land and ocean radiate their energy directly into space.

This works through emitting mostly infrared frequency light for which, just like visible light, the atmosphere is mostly transparent.


However, water can also transfer heat to the atmosphere via conduction, evaporation, etc. I don't know the comparative rates, but there is certainly a lot of evaporation.

It takes about 80 watts to heat up a liter of water 1° celsius. Global data centers used about 4.16 x 10^14 watts last year [1]. Let's assume we still have a lot more data centers to build so bump that number up 10,000X. Assuming every watt of energy is actually converted to heat and we moved every data center on earth underwater it'd raise the temperate of the oceans (1.3x 10^21 liters) about 0.01°C.

This is a very rough calculation and there's obviously nuances but the point is oceans are HUGE and water has a high specific heat. It's much easier to indirectly heat them with greenhouses gases.

Edit: I clearly need to brush up on my physics. Regardless the effect is still miniscule.

[1] https://www.forbes.com/sites/forbestechcouncil/2017/12/15/wh...


This seems to confuse Power with Energy. Watts measure Power, which is a rate of Energy.

The specific heat of water is 4182 Joules/kg and its density is close to 1kg/litre, so 4.128 kJ (Energy) will be required to heat 1 litre of water by 1 degree C.

The temperature rise of the oceans will be complicated to work out. Taking the OPs figures, 4.16x10^14 watts / 1.3x10^21 litres gives 3.2x10^-7 watts per litre of ocean, ie 0.32 microJoules of energy added to the oceans every second. The temperature rise of the oceans will depend on how quickly it can dissipate this heat. What are relevant the heat loss mechanisms? Evaporation just moves the problem to the atmosphere. Conduction just moves it elsewhere on earth. Radiation will shift some of it to space (and some reflected back to the earth), but radiation is a property of the surface of the ocean, not of the bulk.


To clarify: 0.32 microJoules of energy are added to every 1 litre of the ocean every second.

Let's fix the math.

Amount of energy consumed from global data centers: 205 terawatt-hours. [0]

Amount of water in the oceans: 1,386,000,000 (km3)~= 1.386*10^21kg[1]

water has a specific heat capacity of: 4,200 J/kg°C [2]

Energy/HeatCapacity/AmountOfWater = 1.268×10^-7 degrees Celsius

[0]: https://energyinnovation.org/2020/03/17/how-much-energy-do-d... [1]: https://www.usgs.gov/special-topic/water-science-school/scie... [2]: https://www.bbc.co.uk/bitesize/guides/z2gjtv4/revision/5


>It takes about 80 watts to heat up a liter of water 1° celsius

You're confusing power and energy here.


The formula for calculating this is: Q = c_p * d * V * (ΔT). Assuming an isobaric specific heat capacity of 4.18 J/(g K) and a density of 1.00 g/cm^3, heating a liter of water 1.00 K would take 4.18 kJ.

Giving OP's number for the volume of water in the ocean, heating the ocean 0.01 K would be 5.4 * 10^22 J. Assuming 365 days in a year, that's 1.7 * 10^15 W, an order of magnitude less than OPs figure for datacenter power-usage.


> an order of magnitude less than OPs figure for datacenter power-usage

You mean the figure that OP first increased by 4 orders of magnitude, right?


> It takes about 80 watts to heat up a liter of water 1° celsius.

1 calorie is by definition the energy requires to heat 1 gram of water 1°C. In practice 1ml.

So ~1kcal to heat 1 liter of water 1°C

1 kcal is ~1.163 watt-hours.


You are making the assumption that moving data centers underwater would also come along with a 10,000-fold increase in the number of data centers? That seems like an unnecessarily hyperbolic approach.

That would mean going from 8.4 million data centers to 840 billion data centers. I mean, come on...

But hey, let's roll with it. What would be the effect on the environment if we literally had 10,000 times more data centers ABOVE ground, for comparison? After all, you are neglecting to take into account the fact that a HUGE amount of the wattage used by existing data centers has been cooling via HVAC, and by switching to underwater cooling their overall watt usage would theoretically drop a good bit due to the efficiency gains.


I think you are arguing for the same thing your parent comment is? That comment is trying to show that even in the absolute worst case of no efficiency gains and 10,000x increase in data centers the impact that using the oceans to cool them would have on the ocean is absolutely minuscule.

Sounds like they're just using pessimistic numbers to make the point. Even if you multiply the existing datacenter footprint by 4 orders of magnitude, it still only results in a 0.01C raise in temperature.

I believe the GP's point is that even with a 10,000x increase in data center energy consumption, the effect on ocean temperature is tiny.

> Assuming every watt of energy is actually converted to heat and we moved every data center on earth underwater it'd raise the temperate of the oceans (1.3x 10^21 liters) about 0.01°C.

Also assuming we have coated the oceans with a perfect thermal insulator. Otherwise this heat would also radiate away, a small fraction going to the atmosphere and a lot of it back into space.


> every watt of energy

Watt is a measure of power not energy. Did you mean Watt-hour? The specific heat of water is about 4.2 kJ/kg/K. One joule is one watt for one second, that is 1 J = 1 W.s. So one W for one hour (3600 seconds) is 3.6 kJ and is enough to raise the temperature of a kg of water by about 0.9 K

You need to brush up on dimensional analysis :-)


"It takes about 80 watts" - Watt is a measure of power, you need energy instead (Joules / Watt hours / etc )

That assumes heat is distributed instantaneously, which it is not. There will definitely be local affects. The question is if those will harm or collapse native habitats.

There's quite a lot of ocean that doesn't hardly have a "native habitat". Most ocean doesn't look like a coral reef, most ocean looks like water sitting on top of dirt.

Unfortunately, the living parts of oceans are also the parts that tend to be easy to get to, and putting stuff farther away will be more expensive. But careful placement may be able to mitigate that.

It is still ultimately a 3D environment where the vast, vast bulk of the 3D environment is not thriving with heat-sensitive life. Our 2D surface intuition misleads us here. Our mental images of the ocean are of the exceptional locations, not the common ones.


Our 2D intuition leads back on track though, since 2D (or really 1D, or whatever fractal dumention represents thecoastliine) is almost certainly where data centers will get built, not on the middle of the pacific.

3D is relevant to the heat dissipation, and floating submerged data centers doesn't sound so difficult, at least, once you have the idea of a data center not on land. Fortunately, you don't have to go to the middle of the pacific to get away from the living parts of the ocean. You just want to get off the continental shelves, which is easier in some places than others.

Conflating power and energy, and not considering the amount of heat that would be dissipated away, completely destroy any meaning this math would have.

Another question to ask is what the effect of pumping these data centers full of cold air is, and how the excess heat from that may be affecting the atmosphere.

Not to discredit your point or anything, it's a very good one. I think we just also have to answer it in conjunction with what we are currently doing. Maybe we can sink some data centers and have little to no impact while also realizing energy and cost savings?


Slightly off-topic, but this is how I feel when people discuss the risks of nuclear power. You can talk all you want about the disasterous consequences of a reactor meltdown, but what about the disastrous consequences of sticking with fossil fuels?

I agree that switching to wind and solar would be better, but it's not obvious whether we can switch without transforming the world economy in ways people seem unwilling to do. Nuclear reactors can generate much more energy, and I really think it’s time to get behind them.


We need underwater, seaborne, or subsurface nuclear power plants.

Radiation release is contained to within the largest stratum available on the Earth: ocean and rock. Anything avoiding our thin atmosphere is a win.


We need NPT dropped, that’s all it takes I think(given everyone can still be nice after which is a big if)

the "non-proliferation treaty" ?

Yes, nuclear isn’t hard, they’re so strictly regulated because there aren’t other dual use technology that are more controllable, destructive and easily repurposed by sufficiently advanced nations.

My personal opinion is that if we discover a more devastating weapon of some sort, such as kinetic bombardment using bunches of O'Neill space colonies as projectiles, nuclear becomes comparatively benign and that kinds of event can lead to more de-regulation.


From what I remember, the impact was pretty minimal. Water has a huge capacity for heat so the temperature was elevated only when very close to the data center.

the impact was pretty minimal

We used to say that about plastics in the ocean. And landfills. And light pollution. And noise pollution. And space junk. And on and on and on.

Heck, not that many decades ago there were responsible people who thought that air pollution was no big deal because there's plenty of air, and it's just fine if California allows rich car collectors to keep buying leaded gasoline long after other states outlawed it. The impact is "minimal."


This is basic thermodynamics stuff. Looks like energy used in datacenter in 2018 was around 200TWh. If you use all this energy to heat only the mediterranean sea, you will have a change of temperature of 0.00006°C.

I used this calculator https://www.omnicalculator.com/physics/specific-heat


FYI, your comment is wildly misleading. Thermal pollution is real and has a wide research literature.

The worry is not that that the average temperature of the oceans will rise. The worry is that the local temperature of the ocean will rise. This is exactly what happens with other forms of thermal pollution. One example studied for decades was California's San Onofre nuclear power generating station. It has been subject to regulations regarding its thermal pollution into the ocean. They built long pipes so that the temperature increase from the cooling water could heat the ocean gradually over a wide area rather than severely in a narrow area. It's the same principle as a CPU heatsink.

https://en.wikipedia.org/wiki/Thermal_pollution

https://authors.library.caltech.edu/26008/

https://www.worldcat.org/title/ecological-effects-of-thermal...

For what it's worth, I personally found your comment quite distasteful. Instead of trying to understand what you didn't know, you knocked down a straw man and pretended like it was obvious and backed by science.


Thank you for your insightful comments, I stand corrected. I didn't think of the local effects.

This reads like you have not done any calculation to determine whether the entire ocean can be heated by any plausible number of datacenters or not.

What does it matter if the entire ocean is heated? Power plants aren't allowed to dump unlimited quantities of hot water into the ocean. Why is it OK for a data center?

The impact of cooling a building on the surface is definitely higher.

Nothing exists in a vacuum. Of course comparing having a datacentre to not having a datacentre shows that having it has higher ecological impact.

But that wasn't really the option now, was it?


if humans output the entirety of their energy usage into the ocean, it would take ~8000 years to raise the temperature by 1 degree.

There is a difference between understanding of different fields of science. We understand temperature in large bodies pretty well - and we know that it's nearly impossible to sink that many servers to do anything.

I once read that using ZFS was supposed to boil the oceans.

People tend to overestimate and also like to use hyperboles

Yeah, it’s a theoretical limit. You don’t need to boil any ocean to use ZFS.

Reference: >>>

If 64 bits isn't enough, the next logical step is 128 bits. That's enough to survive Moore's Law until I'm dead, and after that, it's not my problem. But it does raise the question: what are the theoretical limits to storage capacity?

Although we'd all like Moore's Law to continue forever, quantum mechanics imposes some fundamental limits on the computation rate and information capacity of any physical device. In particular, it has been shown that 1 kilogram of matter confined to 1 liter of space can perform at most 1051 operations per second on at most 1031 bits of information [see Seth Lloyd, "Ultimate physical limits to computation." Nature 406, 1047-1054 (2000)]. A fully-populated 128-bit storage pool would contain 2128 blocks = 2137 bytes = 2140 bits; therefore the minimum mass required to hold the bits would be (2140 bits) / (1031 bits/kg) = 136 billion kg.

That's a lot of gear.

To operate at the 1031 bits/kg limit, however, the entire mass of the computer must be in the form of pure energy. By E=mc2, the rest energy of 136 billion kg is 1.2x1028 J. The mass of the oceans is about 1.4x1021 kg. It takes about 4,000 J to raise the temperature of 1 kg of water by 1 degree Celcius, and thus about 400,000 J to heat 1 kg of water from freezing to boiling. The latent heat of vaporization adds another 2 million J/kg. Thus the energy required to boil the oceans is about 2.4x106 J/kg \* 1.4x1021 kg = 3.4x1027 J. Thus, fully populating a 128-bit storage pool would, literally, require more energy than boiling the oceans.


This is a non-argument, because it can be used on literally anything, up to lighting one match.

The orders of magnitude are wildly different, even at large server farm scale. According to some estimates, capturing merely ~0.3% of the available energy from the Gulf Stream would be enough to supply all of Florida with electricity [1].

[1] https://en.wikipedia.org/wiki/Marine_current_power#Energy_po...


I'm curious as to whether we'll ever have wind/ocean current energy capture that's sufficient to disturb things enough to cause detrimental (or maybe beneficial) effects.

Supposedly, the recent - apparent - changes in UK climate are due to changes in trade winds through global climate change. There might be small perturbations that we make that inadvertently make for large changes in major wind/current systems?

I'd guess someone has tried to model these things.


I don't have any numbers, but on intuitive level, cooling servers with ocean water directly sounds a lot more efficient and eco-friendly than burning coal in order to generate electricity in order to power AC in order to cool the whole giant room full of air and people in order to cool some servers.

It seems like people are always overestimating the power consumed by this type of datacenter load, and under estimating the thermal power of transportation. This little tube uses 1/4 of a MW. A nuclear submarine has a roughly 50MW(t) reactor, and there are dozens of those prowling around all of the time.

A really huge datacenter might draw 500MW, but that's only the same thermal output as about 7 airliners.

You could submerge every datacenter on earth and no fish is going to notice.


>what is the direct ecological impact of doing this at scale

Great question. Without real data it's hard to know for sure. I do think that it would be somewhat negligible though.

Considering we have heat vents and underwater volcanoes in the ocean that kick out insane amounts of heat, I can't see datacenters having a ton of impact. Will it affect the immediate vicinity, probably. Will it affect the ocean at large, I doubt it. Unless we start sinking exaflops of CPU power into the ocean, I wouldn't worry too much.


Well you could ask same question about lava floating from volcanos to oceans, that much hotter and would have greater impact, or if the sun light would have an ecological impact on the water because it heat it everyday! I don't know what would few degrees increase in temperature around the capsules would do to the environment, keeping in mind that water is changing all the time due to sea current also that temperature is very local I mean it does not extend more than few inches in the water, if you have any knowledge about thermodynamics you would know that the temperature of two bodies should reach an equilibrium and since the ocean is much bigger the effect of the temperature coming from the data centres is negligible, but all this was an experiment and I am sure the ecological factor was taken in account, but still your question was kinda stupid!

Ocean water is already used to cool nuclear and coal power plants. I suspect these would radiate far more heat than a data centre.

I'm surprised msft wouldn't propose to pipe the ocean water in like these power plants before going through the challenge of building under the ocean...


I think they were trying to see if you could get away with 0 power/cooling infrastructure and how well it'd work if you could. Pumping vast amounts of water may be better than running a vast amount of active cooling but neither is as 0 as dumping it in the water and seeing what happens.

I suspect one of Bill Gate's plans is to defend data against solar flares. Deeper the better.

I have the same questions about the temperature. Seems like that might add some extra heat to an already-warming ocean. Especially if this is done at massive scale. I’m not a scientist though!

There is (or was) around a million submarine volcanoes in the ocean according the wikipedia. There are also hydrothermal vents across all of the oceans. The amount of energy to affect the ocean should be colossal (though I didn't do the math). To put it into perspective, we have been pumping crude oil and greenhouse emissions for over a 100 years by billions of people to raise temperatures by a degree or so.

> a body this size radiating 10's of KW of heat

It's kind of literally a drop in the ocean, but I assume it'd be worth doing a bit of research about local disruptions if we intend to drop large deployments in shallow waters.

To shut down or significantly disturb ocean currents, we'd need a lot of these things.


I feel like this is the logical conclusion of any energy conversation.

So what if we make a breakthrough with fusion power? We will all die from heat exhaustion caused by the arse end of AC units bringing about the heat death of the Earth.


Also keep in mind the majority of ocean life is close to the surface so this is also inherently less harmful than a sea side factory dumping heat and waste into the ocean.

I guess the impact would be much less than that from underwater volcanoes and fissure vents that exist already.

Applications are open for YC Winter 2021

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact

Search: