Facebook Testing Lithium-Ion Batteries For Backup Power 41
itwbennett writes Facebook has just started testing lithium-ion batteries as the backup power source for its server racks and plans to roll them out widely next year. Lithium-ion has been too expensive until now, says Matt Corddry, Facebook's director of hardware engineering, but its use in electric cars has changed the economics. It's now more cost effective than the bulky, lead-acid batteries widely used in data centers today.
Li-Ion batteries aren't good for this role (Score:5, Informative)
Lead acid batteries last longest when they are fully charged and kept that way, and discharged infrequently. This makes them excellent for use in standby power situations, where they are almost always topped up ready for the power to go out.
Li-Ion batteries last longest when they are actively used. Keeping a Li-Ion battery fully charged all the time is bad for its longevity; the battery structure breaks down faster at a high state of charge. This is why it is recommended to store Li-Ion batteries half-charged in a cold environment, and why cars like the Tesla Model S normally only charge up to 80% unless you require a "full-range charge" for a road trip. Not topping off to 100% extends battery life.
Maybe Facebook intends to keep the batteries at 80%, but it's hard to believe the economics are going to work in their favor.
Not to mention that lead-acid batteries are mostly water and non-combustible sulfuric acid. A Li-Ion battery fire is 50 times nastier than a lead-acid battery fire, and produces a hell of a lot more noxious gases.
Re: (Score:2)
Re: (Score:3)
Re: (Score:2)
What about energy density?
Wasn't that the driving force for li-ion adoption in computing? It's a 10x difference, according to Wikipedia.
How much space, comparatively would it take for an UPS of each size to keep a given collection of servers running for an hour? If one doubles the physical size of your server farm, and the other increases it by 10% that might be sufficient motivation.
Re:Li-Ion batteries aren't good for this role (Score:4, Interesting)
Lead acid is damned heavy, and offers mediocre density; but if you are just going to shove them in the basement, or are building on the cheapest flat land in the middle of nowhere that you can find, that may not be a problem. However, if cabling costs or resistive losses make 'distributed' UPSes, with fewer big battery banks and more, smaller, battery packs powering individual systems (presumably also cutting DC/AC inverter losses out of the equation and providing DC directly to the motherboard) the superior density and lighter weight of Li-ion is much more attractive.
Re: (Score:1)
On the farm, individual packs would mean you don't lose everything at once during the outage.
I wish our desktop machines offered power supplies with a built in battery, at least for time to shutdown the machine nice and clean. These new all-in-ones are just laptops with a big screen, throw in the battery too.
Re: (Score:2)
Dell XPS 18 (Score:2)
Essentially an all-in-one disguised as a tablet. Works as both.
The i3 version is a dog but the i7 version is quite nice.
Re: (Score:2)
I had a UPS card in my apple IIGS
Re: (Score:2)
You can also find 1u to 4u in-rack UPS systems pretty cheap. I wouldn't power a whole rack with them, but perhaps the critical stuff like storage controllers and the associated fabric.
Re: (Score:2)
I was thinking ultra caps would be good for Datacenter UPS systems. They really only need to work until the backup generators kicking.
Re: (Score:2)
Re: (Score:2)
I assume that Facebook can purchase enough to get all the volume discounts that vendors will offer on 18650s; but I'm genuinely surprised that that's enough, for non-vehicle applications(where weight and bulk are an obvious problem) to beat out the nasty, heavy; but cheap and mature lead acid stuff.
Don't forget that with higher oil prices that shipping costs took a sharp nose up fairly recently. With the latest price drops the lower weight of shipping li-ion vs lead-acid can shift the economics towards li
Re: (Score:2)
They have to be thinking LifePO4's and recharge cyclability. I've been trying to justify moving to lithiums for five years and the economics never even come close unless you count in charge cycles on the lead acids or need the light weight.
Re: (Score:2)
If you want longevity, you can't go wrong with Nickel–iron... I heard about those things lasting 80 years. But, you'll need another building big enough to put them in.
Re: (Score:2)
LiIon is great when you just want 30-60 seconds backup, and is economical there. Knowing Facebook, they likely load balance to charge the racks on off-peak energy and discharge during peak period, even if it is just a minute x15kW per rack for the trivial savings.
"Pure lead" batteries are likely more cost effective, but they are larger and heavier.
Re: (Score:2)
Because nobody at Facebook is an engineer with enough knowledge to be capable of thinking of such things before endeavoring on a scaled test, right?
Oh, and by the way, maximizing the lifespan of a lead acid battery is a wee bit more complicated having them "fully charged and kept that way, and discharged infrequently."
But I'm sure you already know that.
Re: (Score:2)
This is why laptops need a switch for "half" versus "full" charge. If you generally use the laptop docked, you would keep the switch on the "half" setting,, but when you need to use it away from an outlet, you would switch it to the "full" setting and wait for it to charge up before undocking.
Re: (Score:2)
None of which matters if you can get the Li-ion batteries last for as many months as you need them to last and they're priced cheap enough.
Re: (Score:2)
This is not really about the use of Lion batteries. So Facebook is going to change out lead-acid systems that give ~20-30 minutes (MY guess, TFA does not say) to one that gives only 90 seconds for generator start.
That 'dragster' remark is cute but it falls flat with me. There is a whole class of real-world fail opened up here. 90 seconds is scarcely enough time for humans to respond, let alone diagnose and solve a problem. As a critical infrastructure IT admin I'd never want to commit to this. It is an
I thought they were using lithium already (Score:1)
The industry really needs to switch to DC power (Score:1)
One single UPS can cover multiple racks without excessive conversion losses. Putting an individual UPS and a bank of batteries per rack is just plain stupid. Instead of maintaining few large units, you are now maintaining multiple small units. You are now more likely to have an unanticipated failure should the power go out.
Re: (Score:2)
One single UPS can cover multiple racks without excessive conversion losses. Putting an individual UPS and a bank of batteries per rack is just plain stupid. Instead of maintaining few large units, you are now maintaining multiple small units. You are now more likely to have an unanticipated failure should the power go out.
They already mentioned that in the article.... It's better to lose power to a few racks than lose power to 100 racks when your main UPS fails. Your use case may be different but you are not Facebook.
Re: (Score:2)
I used to think the same thing, until I ran the numbers. The simplification in system architecture alone justifies the change, and once you throw in NFPA 70E distributed backup of a local dc bus quickly makes more sense.
It doesn't work for all applications, and you need the IT staff (rather than facilities staff) to make it work, but it can simplify things tremendously.
Re: (Score:1)
One big unit... to go BOOM! and take everything down at once. My evil plan cannot fail... this time
Re: (Score:2)
With one battery per server you can design a custom power supply that's made with it in mind, and you can easily finely manage the battery and implement any scheme. e.g. server immediately throttles down when switching to battery power ; power is drawn from the battery when you spin up hard drives (if that makes sense), battery monitoring and power schemes are maybe a good job to run on a hypervisor.
Re: (Score:2)
Not in my (limited) experience. This was a dispatching center for a large (Class 1) freight railway. For each of about 40 dispatchers there were 5 machines a pair of PCs at the "Desk" plus a trio of PCs in the back room for server/backup server & maintenance workstation. The "ONE BIG UPS" failed 3 times in 4 years, and each time the whole damn railroad was stopped dead for several hours. ( Egon Spengler - That would be bad.)
We put a small (about 2000VA) UPS at each desk and another for e
Re: (Score:1)
Re: (Score:2)
I thought they were already using DC power:
"The battery cabinet is a standalone independent cabinet that provides backup power at 48 volt DC nominal to a pair of triplet racks"
http://www.opencompute.org/pro... [opencompute.org]
http://www.opencompute.org/pro... [opencompute.org]
"Google Uncloaks Once Secret Server" (Score:2)
http://www.cnet.com/news/googl... [cnet.com] seems similar. They claim 99.9% effective utilization through their per-server battery backup system, compared against 95% for a centralize lead-acid UPS based system.
http://hackaday.com/2014/11/11... [hackaday.com] might also have some nuggets. a lead acid battery is going to be heavily de-rated at the energy rates required. lead-acid will likely not have the same charging efficiencies.
holding the batteries around 70% is no big loss for this use case, given that the alternative is shorte
Really? (Score:2, Insightful)
Is the data kept by Facebook worth anything?
This makes no sense to me. (Score:2)
Given then Facebook should have acres of servers, why are their data centers not using a flywheel / diesel combo for power backup. [wikipedia.org]
If you only have a couple of racks of servers, batteries make sense, but they should have thousands and you need something like a generator if your power can be longer than your battery will last. Its not like this is a new and unproven technology.
Is there really any good reason to consider batteries for a large data center?
Re: (Score:1)
Should be clear, I know facebook is using batteries for a short time to serve as the power till the diesel kicks on. But how can you beat a flywheel in-line with the diesel.