I was also surprised at some of the patents awarded. While the claims were very specific, the ideas were very generic, such as mobile payments via cell phone. Take a look for yourself: http://www.google.com/patents?zoom=4&q=inassignee%3Agoog...
If anyone finds the battery/server patent, I would be extremely interested.
"Each UPS includes a battery selectively connectable across a DC bus, and a AC-to-DC rectifier that converts an AC input voltage to a single output voltage on the DC bus. The regulated DC bus voltage may be close to the battery's fully charged voltage..."
Note that the publicly available Gigabyte board you link to is the GA-9IVDT, the name on the board in the photo is GA-9IVDP. Moreover, they look nothing like each other.
I also can't tell where the 5V rail for the hard disks is coming from:
It's quite possible that they got Gigabyte to do a custom design for them, given their volumes. Or, the 5v conversion is done in an extra device between the power supply and the mother board power socket (ATX or whatever it is).
I don't see how they can patent battery per motherboard design. Isn't that how laptops work?
OTOH, this design makes a lot of sense. One of the biggest failure mode for rack-wise failure I've seen is human error from maintenance crews: trip over or unplug the wrong a power cord (I've seen this happen twice already from different companies.) This design completely shields this type of errors.
The battery-per motherboard makes eminent sense. Between economies of scale driving down the price of small AGM and gel-cell lead-acid batteries (1) and the increased efficiency of having the battery on board, it's an obvious win!
(1) - ironically, this is probably because of retail and commercial sales of Uninterruptible Power Supplies!
1) No transmission/conversion loss. It is probably soldered directly to the DC power regulator on the mainboard.
2) The servers are frequently removed for maintenance. Thus, the batteries will also be checked/replaced along with the machines they serve. As a bonus, there is always exactly the right amount of battery capacity in the system.
I agree. Standard UPSes convert AC to DC and back, in order to power the server PSUs and these conversions are inefficient. If the battery is in the server (after the PSU), you can miss out these conversions and save a lot of power.
Could also be to do with management, particularly switching - switching large DC currents is problematic (i.e. it arcs). It's easier to manage on a smaller scale.
With the one battery per server, you eliminate one moving part. Well not moving, but you leave out the step of going from dc power to ac power. With their setup, you take 220 vac (note the label on the power supply) to the server and to dc. Any conversion step costs heat.
See how the hard drives are mounted back-to-back? I'm guessing that is designed to cancel the mechanical noise by putting the drives 180deg out of phase from each other, extending the life of the drives.
I believe that is patented (by Avid?) and that might be why Google was cagey about releasing specs.
Had the read the article carefully to see if i wasn't "getting" the joke. It's legit, and very cool that they were so innovative with their data centers.
The battery integration is quite interesting, although I'd bet they are using it for purposes other than "having to buy one big UPS."
One benefit I see to it is high availability, since the site UPS going out won't take down the servers, since they all have their own. Plus, I'd bet it makes the guys that work on the power buss sleep sounder at night, since its easier to work on without worrying about taking out the whole datacenter.
Yes, there have been a few publicized incidents at large hosting centers where the central UPS either failed or didn't kick in when it was supposed to. By putting the battery in the server it eliminates that single point of failure.
I've been in data centers hosting Google systems. This picture looks very similar if not identical to what I've seen, so I'd say it is most definitely real.
Based on their white paper on MTBF of hard disks, I'd guess that they avoid favoring any single vendor to lower the chance of a bad batch of drives all failing at once.
Battery pack is a good idea. I have a UPS that is just about dead (~1 minute of backup, but I keep it for spikes and drops). I thought it would be great if the PSU had a built-in UPS. The space between the PSU and drive bay would be a perfect spot, no need for the heavy switching hardware since the PSU can deliver the 12 or 13.6 volts needed for the battery to charge.
What battery chemistry are they using? It looks like a lead-acid package. Those are cheap, safe, and reliable. But they have very low energy density and there are disposal issues.
it looks like two of them may fit side-by-side, like this they get to use the big ram sticks and a cheap psu, as well as fans that will last a lot longer than those pesky 1u fans.
They may also use 'deep' racks to the machines have the PSU's in the middle and the fronts on the facing sides of the cabinets, one front, one back. That would make it twice as efficient as 1U, it would also explain why the cables go out the front.
I've seen them two deep side by side. They had custom racks that reminded me of those racks of trays of food you'd see in large production kitchens. There were four systems per level/tray.
I switched back and forth between 1U and 2U setups for the same motherboard, and I could run my system in higher ambient temperature with less airflow when using a 2U heatsink.
From the patent-looking picture in the article, it looks like the racks are 1-deep, so I don't think they're using 2-deep racks.
I was also surprised at some of the patents awarded. While the claims were very specific, the ideas were very generic, such as mobile payments via cell phone. Take a look for yourself: http://www.google.com/patents?zoom=4&q=inassignee%3Agoog...
If anyone finds the battery/server patent, I would be extremely interested.