Slashdot Mirror


Google's Addiction to Cheap Electricity

Googling Yourself writes "Harpers magazine has published a blueprint of Google's new data center at The Dalles, Oregon where they will be tapping into some of the cheapest electricity in North America. Although the plans show three 68,680-square-foot storage buildings, only two of the buildings have been constructed so far. Based on a projected industry standard of 500 watts per square foot, the Dalles plant can be expected to use 103 megawatts of electricity. Google's server farm represents a new phase in the transformation of the Columbia River over the past half-century. Across the street from the Google data center is an example the last generation of high energy consumers; Microsoft, Yahoo, and Ask.com are also planning data centers on the Columbia River."

20 of 254 comments (clear)

  1. Saskatchewan by corychristison · · Score: 2, Informative

    Come put one in Saskatchewan. It would benefit our economy and our (commercial) power rate is 0.0845-ish per kWh.

    1. Re:Saskatchewan by corychristison · · Score: 2, Informative

      Haha.

      You use 'eh?' pretty much when you please. "How's it going, eh?" is a prime example.

      I pronounce schedule as "skedual." Others in the area pronounce it as "shedule." Both are widely accepted.

      As for the Hockey vs Football remark, I'm not sure. Considering the two dollars are pretty close at the moment, I'd have to say that would be a pretty good metric to go on.

  2. Re:For Profit Company is Cost Conscious by Russ+Nelson · · Score: 4, Informative

    Well, yes, and it's a strange point of view to say that a company is "addicted" to one of its inputs. One may as well say that Google is addicted to CPUS, or to buildings, or to fiber optic cables, or to people.

    --
    Don't piss off The Angry Economist
  3. Re:I thought they were going green? by calebt3 · · Score: 2, Informative

    The water in the glass next to me isn't falling...
    ;-}

  4. Re:I dont understand this "cheap electricity" thin by hidden · · Score: 4, Informative

    Transporting large amounts of power still costs money... all those 320kV lines? Those use large amounts of copper ($$), they have to be mantained ($$)... There are some lines in place, yes, but the more power you send farther, the more cable you have to run, and up goes the cost of providing the power. That cost then gets passed on to the consumer, in the form of not-so-cheap-any-more electricity.

  5. As an Oregonian... by rampant+mac · · Score: 4, Informative

    I live in Portland and this is the first I've heard of various tech companies building along the Columbia. It kind of sucks to see Oregon becoming more popular - something like 95% of the state lives west of Portland. I don't want to see the state becoming like California.

    So if you're thinking of moving to Oregon, remember: It rains here ALL THE TIME. There's hippies everywhere. Nearly half the women in Portland are lesbians too!

    Actually, I didn't make that last line up. :(

    *sigh* Ever our governor once said "Oregon: a nice place to visit, but please don't stay."

    --
    I like big butts and I cannot lie.
  6. The Columbia flows from BC thru WA/ID and WA/OR by WillAffleckUW · · Score: 2, Informative

    I don't know for sure about the other states/province but they've been trying to get tax writeoffs (aka bribes of tax dollars) to build more server farms here in Washington State.

    And supposedly, the other states - Oregon and Idaho.

    British Columbia, which has most of the dams (and is building two Columbia River treaty dams now near Revelstoke BC and Trail BC) provides most of the power, so I presume they may also be included in these attempts to get tax subsidies plus cheap power.

    --
    -- Tigger warning: This post may contain tiggers! --
  7. Re:I dont understand this "cheap electricity" thin by Jeffrey+Baker · · Score: 2, Informative

    Not from around here, I take it? We have an entire department of uneconomical projects. We call it the Bureau of Reclamation. Their job is to build vast, expensive capital projects and subsequently give the benefits to select landowners for a pittance. For example, the Bureau will build a dam and sell the electricity to nearby farmers well below the market price of electricity, sometimes even below the cost of generation. The Bureau will then also sell the water from the dam to the same farmers at a fixed cost in perpetuity. On top of all that, a typical Bureau farm project grows a needless crop which the federal government subsidizes. You can find plenty of farms in the western USA where the farmers get electricity at $0.05 per kilowatt hour and water at $9 per acre-foot and are growing federally subsidized crops such as cotton. In the vast majority of Bureau of Reclamation projects the initial capital costs have never been paid by the farmers who benefit.

  8. 500 w/sf "industry standard"??? by jhw539 · · Score: 2, Informative
    I'm not sure where they're coming up with that 500 W/sf number, even for 2011. The numbers I've been seeing and designing to are more like 200-300 W/sf. Above that, things get awfully tricky to cool using just air. Take a rack at 500 W/sf. The rack footprint with access aisles (included in the industry standard definition of W/sf is the access space) is at least double the rack footprint plus space for pdus and support, so call it 20 square feet. So you need to move out 10 kW with air. Assuming a supply air temperature of 70F (if you're supplying lower, you're pissing away energy and need to get modern equipment or update your education*) and a temperature rise of 30F across the computer, you need to move (forgive the conversion to English units, my mental math doesn't convert)

    8000 Watts = 27,300 btu/hr

    (34,100 btu/hr) / 1.08 / 30 = 1050 CFM per rack

    And that is assuming that you have an airtight seal between the hot and cold aisles (which you should be pretty close to if you're going that dense)(or you should fire your mechanical designer). Unless you increase the footprint or get crazy and supply from above and below (possible but costly), you're supplying that air through a cross-section of 4sf at a diffuser velocity of 250 fpm. That's not a datacenter, it's a windtunnel.

    Now, I am going off a bit half cocked here. You could indeed get up to 500 W/sf if the hardware is designed to be cooled, probably by water but possibly by high delta-t air. And the potential for entirely passive cooling of that level of heat is there. I'm just a little sensitive to people tossing out numbers that are patently silly. If they'd at least use an intelligent fan supply algorithm on the recircs (and please god, don't use those constant flow CRACs) it wouldn't waste quite as much power.

    As a side note, if google is spending 0.5 Watts to cool per 1 Watt of cpu heat (a power usage effectiveness of 1.5) in their new Portland facility, they need to put some of their eggheads** to work trolling the current datacenter efficient design literature. I know their solar cell powered campus is sexy, but they could save an order of magnitude more energy for a tenth the cost by getting their datacenter up to snuff (the obvious freecooling, the barely-justifiable-but-sexy adsorption chillers, and the esoteric thermal stack harvesting to offset fan power for starters)(a rigid hot aisle/cold aisle design is, of course, assumed just to get a 300W/sf datacenter that won't cook the top slots with recirc).

    * OK, if you can make cold air for free via free cooling, go for it but I expect you to do the work on the fan controls to actually harvest a few kW's for the added humidity control complexity. ** Based on a presentation I saw at ASHRAE, Google does have at least one sharp tack in the drawer, so I suspect this is a matter of a Harper's 'expert' making up numbers.

    1. Re:500 w/sf "industry standard"??? by wwwillem · · Score: 3, Informative

      Let's try and do some math. A standard industry 1U server is nowadays using as much as 500 Watts. OK, for arguments sake, let's half that, utilization is not always 100%. Then we have 40 servers in a rack, so that adds up to 10 kW per rack.

      Now a rack is 2x3 feet, but you need space in front and at the back, so lets take 2 feet wide (that doesn't change) and 10 feet deep, a total of 20 sq.ft. In which case we get to a power consumption of exactly 500 Watt/Sq.Ft. Most datacenters will not have this model of 40 1U servers in a rack running at full blast. But Google probably is one of those that do exactly that.

      Once I was in a co-location datacenter where one of the cages was occupied by google. That was still the time when they built their own servers, 4 motherboards in a 1U tray, 144 MBs in a rack. In this case / cage :) I was looking at roughly 20 racks of servers. And the heat that came out of that row, man oh man, it was pretty intense.....

      --
      Browsers shouldn't have a back button!! It's all about going forward...
    2. Re:500 w/sf "industry standard"??? by Anonymous Coward · · Score: 2, Informative
      Have you ever *measured* the power being consumed by a datacenter? A mere 100 W/sf will give off pretty intense heat, what is the *airflow*. How may square feet of duct or unobstructed underfloor was dedicated per rack? Could you stand upright in the supply ducting? The scale of honest 500 W/sf is very apparent. I don't use less than a 3' raised floor or a 24" main per row just for 300 W/sf (unless you're talking a little 500 sf university playroom, which essentially becomes part of the air handler unit). And if google has a rack of 40 1U servers then they have to have the PDU's, switches and patch panels located somewhere in the datacenter footprint that brings down the actual design W/sf. OK, there are ways to get those out of the datacenter footprint, but that opens up arguments over the definition of datacenter footprint. I've measured a couple dozen large datacenters and colos - 500 W/sf doesn't happen unless you discount access space and move all supporting equipment including patch panels outside your "datacenter" footprint.

      Yes a rack of 1U servers is "intense heat." No, an entire large datacenter does not run at 500 W/sf in the forseeable future. Maybe when the helium cooled quantum machines come in, but those have internal cooling loops that can dump to a process cooling loop last I looked (and I still have trouble believing they are honestly aiming to market them in just a couple years, albeit for very limited and specialized applications - public key is safe for a while yet...).

      But as a final note, if they go to liquid cooling (as some players are quietly doing building coils into the racks or similar) rules change. It's still awful tough to have that many 1U servers without the low density traffic routing gear and PDUs diluting your W/sf, but maybe they're getting custom 12' high racks just for the hell of it or something. Sure doesn't make engineering sense - this isn't a cleanroom, square footage is not that expensive for a datacenter.

    3. Re:500 w/sf "industry standard"??? by jhw539 · · Score: 2, Informative

      Doh! I hate it when I accidentally post anon... I'm the measurement worshiper above. In god we trust, all others bring data...

  9. Re:I guess Google hasn't gotten the memo. by karl+marx+is+my+hero · · Score: 2, Informative

    That theory is nonsense as per the paper's own admission. The theory is premised on the fact that there will be no more exponential grow in the nuclear and hydroelectric industries, which is a patently absurd assumption. The main focus of the paper is concerned with peak oil and coal.

  10. Re:Wind power? by The+Second+Horseman · · Score: 3, Informative

    No, I got it. It would be hard not to - I think I've heard a "hot air in Washington DC" joke about, oh, four or five times a week for the past 35 years. But the power situation in the DC area is particularly bad, so I figured I'd point out that it's a lousy place to try build a data center.

  11. Re:I dont understand this "cheap electricity" thin by evanbd · · Score: 3, Informative

    Actually, long-haul lines are all alumium these days. The resistivity is slightly higher, so the lines are thicker, but the aluminum costs less and weighs less for the same load capacity as the copper.

    Your point is still completely correct, though.

  12. Re:Power and Cooling - the top DataCenter expenses by Anonymous Coward · · Score: 1, Informative

    The Niagara/Buffalo systems feed their densely populated surroundings. There are even nuclear stations in the area to meet demand.

    As for cold air, that's seasonal. Summers are very hot in that region, with regular requests that people ease off the air conditioning during the worst of it. So, yeah, not a good place for a big datacenter actually.

    (For non-locals, Toronto is so miserably hot in summer that even your friends from *Indonesia* complain about it. But the rest of the time we live in igloos, you bet.)

  13. Re:Tax Breaks by TubeSteak · · Score: 3, Informative

    This line of reasoning always confuses me. How is giving someone a tax break the same as giving them a subsidy? This isn't a difficult concept. Subsidies are not only cash.
    Subsidies are any form of government granted (financial) benefit.

    http://en.wikipedia.org/wiki/Subsidy#Tax_Subsidy
    Feel free to read about the other types of subsidies.

    In this case, [company] gets to use the full range of government services without paying the same taxes like everyone else.

    In TFA I linked, Yahoo and Microsoft are threatening to build their datacenters somewhere else unless they get (amongst other things) a specific exemption from the 6.5% sales tax on their purchases for the datacenters, because "the [Washington] state Department of Revenue recently determined that the server farms aren't eligible for an existing tax exemption for rural manufacturers".
    --
    [Fuck Beta]
    o0t!
  14. Re:Microsoft? by chromatic · · Score: 3, Informative

    Just out of curiosity, what kind of pollution do you expect a data center to produce?

    Heat. Living things grow in the Columbia River.

  15. Re:Power and Cooling - the top DataCenter expenses by astro-g · · Score: 2, Informative

    yes, but the outside temperature drops at night, making the air cons job a lot easier.

  16. Re:They Could Buy Different Servers by Anonymous Coward · · Score: 1, Informative

    Mainframes, for example? Google's current deployment architecture is simply very energy- (and space-) intensive. If you've got knowledge of the internal architecture of Google's systems, the rest of the industry would like to know. Please either cite some sort of reference so we can all read it, or admit that it is just speculation (much like TFA).

    Different servers are optimized for different types of computing, and Google does a lot of what mainframes do best. A jack-of-all-trades machine will always be less efficient than one built to suit the application. Why?... because the do-everything machine will need all the parts, while a do-one-task machine can drop the parts that it won't use. Why have graphics or a PCI bus on a machine that doesn't need it?; Why put a disk on a machine which only does CPU calculations, and why put extra RAM on a machine that only reads its disk? Specialization can be annoying for applications, but inefficient it is not (assuming a large enough economy of scale that you can design/build the customized systems of course). Then again, this is all wild speculation anyway, unless you can come up with that reference.

    Maybe Google will figure that out before trashing the environment any more. With 15k employees and an army of PhDs, I would guess that a least a few are focused on efficiency. If a company can hire a person who can do calculations which save that company millions of dollars in hardware or energy, they do it. Even the most daft of companies still know that.

    P.S., if you can do a greener websearch on mainframes, by all means go for it. I'd be the first to start using that if it works, but don't expect me to bet on your idea.