Microsoft declares its underwater data center test was a success

Underwater pods can reduce latency by moving cloud services closer to customers.

Read the whole story

 

[SixDegrees]

While this is great for cooling, especially up there, I can't see any other meaningful benefits to doing this.

There's so many factors which could take the thing offline, stormy weather, shipping, fishermen, leaks.

Cool idea but just doesn't seem worth the hassle

You should consider reading the article before commenting.

 

[SixDegrees]

The failure rate of one-eighth that of normal servers is impressive, and as the article notes is probably due in large part to the nitrogen atmosphere in the container, which greatly reduces oxidation. It's not all about cooling.

 

[SixDegrees]

Maintenance isn't the problem many are making it out to be. Server farms have enormous redundancy built in; it isn't necessary to rush in and replace a single faulty hard drive. More likely units like these would just be left down there until some significant percentage of failures occurred, then hauled up for repair and re-sunk.

The eight-times better lifetime of equipment found in these trials makes maintenance even less of a concern.

 

[SixDegrees]

Undersea pods that could be placed near major undersea cables - Sounds like an NSA dream come true.

Meh. The Navy has had that capability for many decades. Commercial deployment doesn't bring anything new to the table.

 

[SixDegrees]

Chips last a lot longer if they're kept cool - all the failure mechanisms go up non-linearly with temperature. Running cooler might also be contributing to the better reliability here. If these pods could be placed in even deeper water, they'll be cooler still.

But I wonder about the effect on sea life. You don't want to disrupt the local ecology and probably the fishing by putting these warm pods all over the sea floor. If this idea is as attractive as it sounds, there might be thousands of them off the coasts of Seattle, New York, and Shanghai. Maybe that just means they have to be spaced out more.

As chips get more efficient, they produce less heat per unit of computational work. But that also means that more of them can be packed into these pods. Maybe if Moore's Law beats the increasing demands on servers, the pods can run cooler, but it could just as well go the other way.

Read about Hydrothermal vents. If anything, the pods will help.

TL;DR - heat is good down in the cold and dark. Also read the above posts about how absolutely insignificant this heat source would be. Not even a rounding error.

After a couple of years, I'm actually surprised at how little marine life took up residence on the container. I would have expected more encrustation.

 

[SixDegrees]

Maybe I’m wrong in assuming the more naval aspects of all this were a solved problem. The Gulf of Mexico at least is littered with seafloor vessels doing all kinds of things. I figure we know pretty well how to put tanks underwater.

For that matter, I can’t help wondering what offshore customers Microsoft might already have an eye on for these. What kind of geology can you do when your DC isn’t at the other end of a satellite link, but instead parked right under your rig?

But what’s fascinating for me is what life must be like for a server in one of these things. Five pages deep into the thread, we’re still arguing about thermodynamics and heating buildings on land (???). Heat is an interesting issue, but these aren’t underwater toasters. They’re complex systems adapted to a uniquely harsh and remote environment. I would be interested in knowing more about the design and operational considerations, and I’d love to know more about what MS learned about those aspects.

Or is all that a solved problem too?

Meh. The computer doesn't know it's in a hostile environment. It actually likes the nitrogen rather than oxygenated air apparently. The question asked before was whether these cases were pressurized to match the depth or held at one atmosphere. I would be interested to know the answer to that - especially as relates to the performance of spinning rust hard drives.

As for the rest, so long as a computer in its rack receives it electricity, has its heat removed and has a data connection to the outside world, it doesn't care where it is. It's not like there's a vibration or orientation issue with being submerged.

The article describes the back-up power concern for operating off a renewable power generation system. Optical links underwater are no big thing. All the sub-oceanic data cables have repeaters along their lengths. And the heat removal was addressed partially in the article. The ocean was used as a heat sink. Personally, I'd like to know if there was a coolant loop to efficiently couple the servers to the shell or if they just relied on fans but that would couple into the operating pressure discussion as well.

Another point in favor of this environment, mentioned in the article: you don't have people wandering around bumping into things, spilling coffee on things, plugging and unplugging things, and generally acting as biological sources of chaos.

My software/hardware lab would almost never have any problems if I locked all the people who use it out of it.

 

[SixDegrees]

Keep in mind that serious datacenter operators (aka cloud providers) have been working to get their power usage efficiency (PUE = ratio of total datacenter power consumption to computer-hardware-only power consumption) as close to 1 as possible for quite some time now. Google currently reports a fleet average of 1.11 https://www.google.com/about/datacenters/efficiency/

So, compared to these highly-optimized surface datacenters, there isn't that much power to save by reducing cooling consumption to zero thanks to full seawater immersion. I mean, avoiding a ~10% overhead is always nice, but nowhere near game-changing really. The reliability increase may indeed very possibly come from running hardware sub-10°C rather than in a >20°C ambient air.

Regarding EMP resistance, clearly with that much (decently conductive) seawater above and a relatively thick steel shell these pods are well hardened against radiated effects. As for pulses conducted down the (most likely steel-armored coaxial) power cable, it's difficult to say without knowing the exact cable impedance characteristics & pulse shape... not something I have much experience with so I'll refrain on commenting further.

As noted in the article, though, those aren't major considerations. The rather stunning eight-fold improvement in failure rate, and the relatively cheap real estate that's still close to major populations, factor in a lot more heavily.

 

[SixDegrees]

Interesting idea. Solves reliability issues and provides inexpensive cooling for those power-hungry servers.

But the fiber optic cables connecting the servers to the mainland may be vulnerable to sabotage.

Maybe. But they're cheap, and easy to repair. There are literal tons of undersea cables out there, and although snags occur, we know how to deal with them.