Microsoft Sinks Data Centre Off Orkney To Test Energy Efficiency (bbc.co.uk)

An anonymous reader writes: Microsoft has sunk a data centre in the sea off Orkney to investigate whether it can boost energy efficiency. The data centre, a white cylinder containing computers, could sit on the sea floor for up to five years. An undersea cable brings the data centre power and takes its data to the shore and the wider internet -- but if the computers onboard break, they cannot be repaired. The operation to sink the Orkney data centre has been an expensive multinational affair. The cylinder was built in France by a shipbuilding company, Naval, loaded with its servers and then sailed from Brittany to Stromness in Orkney. There, another partner, the European Marine Energy Centre (EMEC), provided help including the undersea cable linking the centre to the shore. "This is a crazy experiment that I hope will turn into reality" said Ben Cutler, who is in charge of what Microsoft has dubbed Project Natick. "But this is a research project right now -- and one reason we do different types of research into data centres is to learn what makes sense before we decide to take it to a larger scale."

Comments

Even better

[Big Bipper]

Why not build data centers into the bases of offshore wind turbines. You would still get the, for all practical purposes, infinite heat sink of the ocean or large lake, cheap energy ( most of the time when the wind blows ), access for repair, and the data cable could be laid with the power cable from the turbine. Everybody wins.

Re:Even better

[whoever57]

the point of this project isn't to cool a datacenter using the ocean as a heat sink. It's to build a self-contained datacenter which can be customized to order, is easily shipped (fits on the back of a truck), then deployed at the bottom of the ocean where it operates for years without human intervention.

So why put it at the bottom of the ocean? Why not in a container near the ocean? In most of the places where this will be deployed, land is cheap. The Microsoft document you point to isn't very clear on the advantages, but it does include this:
"The world's oceans at depth are consistently cold, offering ready and free access to cooling, which is one of the biggest costs for land-based datacenters."
It does also mention the idea of co-locating with a turbine, but I think the statement above suggests that energy efficiency is one of the primary goals.

Re: Even better

[mysidia]

Replacing a failed HDD is going to be a real bitch?

At our scale, messing with individual HDD's isn't economical... we'll just replace the entire datacenter when necessary.

or... The admin will enter a command to blow electronic fuses on the failed HDD's data and power lines to ensure it fully disconnects from the backplane, then copy and paste a new License Key received from the storage manufacturer to activate one of the pre-stocked replacement bays. Either that, or the storage will be All Solid-State with an automatic failover system and artificially reduced capacity to provide increased redundancy. Ship it with 4X the modules required for the unit to work over its service life within the expected failure rates, and there will be no need to replace anything.

Re:Even better

[drinkypoo]

Huh? So? That's a very minor engineering problem compared to most of the other issues.

It really isn't. What happens when 1.21 GW passes through your server while the HDD spins up past 8,800 RPM?

Not a long term solution.

[Going_Digital]

This is clearly not a long term solution, the oceans are warming and that is already causing concerns. Sticking a bunch of immersion heaters in the ocean is not exactly going to help.

Re:Not a long term solution.

[Oswald McWeany]

This is clearly not a long term solution, the oceans are warming and that is already causing concerns. Sticking a bunch of immersion heaters in the ocean is not exactly going to help.

Against the vastness of the ocean; underwater data centers are going to make no statistical difference to the temperature of the ocean.
Even if temperatures in the environment raise by the forecasted 2C- that's not going to drastically impact the cooling ability of the ocean either.

Of course, it would be even better if the data center was in low orbit.

Re:Not a long term solution.

[MightyYar]

Alright, I need help here. Sarcasm or not?

Re:Not a long term solution.

[cellocgw]

This is clearly not a long term solution, the oceans are warming and that is already causing concerns. Sticking a bunch of immersion heaters in the ocean is not exactly going to help.

Even worse, what if they build one of these near R'lyeh ? He might wake up from the warmth.

Re:Not a long term solution.

[barc0001]

The warming effect of these "immersion heaters" on the oceans is insignificant even if all datacenters in the world were already sunk in the ocean. Every second of every day, the Earth is hit by 1.74 x 10^17 watts of energy from the sun. So even 100 gigawatts of "immersion heaters" in the ocean would be less than 0.001% of the heating effect that the oceans already get from solar radiation.

The real benefit is that 15-20% reduction in electricity use for cooling, especially if that's 15-20% less fossil fuel generated electricity, which contributes to greenhouse gases.

Re:Not a long term solution.

[Memnos]

Ultimately any energy savings to be had by sinking datacenters probably isn't worth the added infrastructure cost. Adding "marine" to anything tends to make the cost 3x what it would otherwise.

They could offset the extra cost with extra "value" - instead of just adding "marine" they could add "marine blockchain".

Re:A lot of sunken capital

[Tablizer]

It's missing a key device diver.

Repairs

[darkain]

Servers in data centers now are hardly ever repaired. Why spend the money? When you're running 10,000 servers and 1 breaks? What is the cost of that single unit vs the time to troubleshoot and solve the issue? All of the software and data is designed to be redundant anyways nowadays. The data will just be shifted around, and the processing load shifted as well. So having no access to fix things is mostly a moot issue. And 5 years? Thats about the max length of a server in a data center as it stands right now as it is. Overall, this sounds like a good scenario!

Re:Repairs

[Anonymous Coward]

Servers in data centers now are hardly ever repaired. Why spend the money? When you're running 10,000 servers and 1 breaks? What is the cost of that single unit vs the time to troubleshoot and solve the issue?

Do you have any idea how cheap it is to hire a warm body, even if they got the super duper awesome A+ certification, to maintain thousands and thousands of servers? At a university, we pay ours $11/hour for less than ten hours per week to service a thousand servers. He pays for himself many times over every week compared to letting even a single server go unrepaired.

Let's say you can't get labor that cheap and your version of cheap labor costs $30/hour with benefits (keeping in mind that this is just slightly more than a menial job to diagnose and replace DIMMs, motherboards, hard drives, etc so $30 is too much IMO). For 1000 servers you probably need much less than ten hours in a typical environment, but let's round up to ten. That's $300/week. If he fixes just one $5000 server per week, he has paid for himself even if he's on Facebook during times when nothing is broken. We are an HPC center and stress these servers to the max, so we do have multiple parts failures in a typical week.

What morons just let servers die and throw them away when all you need is a replacement DIMM? I realize that some people are stupid or just haven't done that math but wow, that sounds idiotic to just throw away servers when it's *much* cheaper to fix them. (And yes, I realize that having to dive down to the data center wouldn't be a cheap or viable solution, so my math doesn't apply in the case described by this article).

Re:Repairs

[Asgard]

How many more machines could you fit in a given space if you didn't have to handle a human gaining non-disruptive access to a particular machine? At scale people start looking at a rack as the unit-of-replacement, maybe even a row. In this scenario the entire sinkable unit becomes the unit of failure.

This isn't particularly new, Sun had Project BlackBox https://en.wikipedia.org/wiki/... in 2008.

Re:Repairs

[angel'o'sphere]

I realize that some people are stupid or just haven't done that math but wow, that sounds idiotic to just throw away servers when it's *much* cheaper to fix them.
Plenty of big data centers do that.
Google uses to have cargo containers fitted out with racks of computers, drives and UPSs.
They placed them somewhere and let them run, they never repaired anything, when the whole container dropped below 50% capacity they dismantled it and threw it away (because that took 3 or 4 years and the then actual tech was more advanced)
No idea if they still follow that policy.

Leave the sea alone?

[thePsychologist]

There are so many lifeforms in the ocean that people rarely see, so if this sinking data center idea takes off, the massive number of sunk data centers could affect these lifeforms and no one will watch out for them.

Yes, it might reduce the CO2 and barely warm up the sea, but there are other aspects to balancing this equation than just this.

Re:A lot of sunken capital

[tlhIngan]

Shark Week's coming soon!
Azure vs. the Great White
Can Megalodon do Office365?
Teaching Sharks to Surf (the Internet)

Re:Watercooling

[cellocgw]

It is all about watercooling, man...

So, since they're cooling PC boards, is it ....

waterboarding?

Why does it need to be deep?

[FryingLizard]

Why sink it? That shit sounds expensive. The only thing you're after here is free cooling; why can't it be on the shoreline, or say 50ft offshore? Stick it in a concrete bunker if you like; run a water pump or arrange for natural sea currents to do the work. It's good enough for nuclear power stations.
This sounds like a toy project.

Re:Why does it need to be deep?

[ElizabethGreene]

Or an attempt to be outside the dominion of a country that does /not/ have data protection laws.

The sword cuts on both sides.

"Microsoft sinks data center"

[Locke2005]

I'm sure this is by no means the first data center sunk by Microsoft!

Re:Can’t sweep heat under the rug.

[iggymanz]

hahaha, you know all the power consumed by human civilization, if converted to heat and dumped in the ocean, woul not make an iota of difference compared to the sun. Even the natural variation in solar output totally dwarfs the heat output of mankind. We do NOT have a mans-waste-heat-warming-the-earth problem.

Pollutions making gases that trap a bit more SOLAR heat, pollution darkening ice and snow to trap more SOLAR heat...yes, those might be a problem.

Re:Watercooling

[Joce640k]

Never mind the puns, how long before somebody steals it?

A bunch of Somalian fishermen with a supply of large inflatable bags will have that thing off the sea floor in no time.

Re:A lot of sunken capital

[sjames]

Microsoft has been sinking data centers since NT.

Re:Watercooling

[Cederic]

You must be American. Somalia is a 9000km boat ride from Orkney, assuming they use the Suez canal and don't get arrested or sunk on the way.

I'm thinking there are easier ways for them to earn a living.

Shit fuel costs and canal fees will be higher than the value of whatever they dredge up.