digg

Facebook Connect Join Digg About
See the original image at news.cnet.com

Google uncloaks once-secret server

news.cnet.com — Unusually, the search giant designs its own servers. For the first time, Google unveils one publicly, showing a surprise built-in battery.

Who dugg this? Made popular Apr 3, 2009

Digg DudeWhat is Digg?

Digg is a place for people to discover and share content from anywhere on the web. Digg surfaces the best stuff as submitted and voted on by the community. Take the Tour to learn more.

252 Comments

show profanity settings
expand all
  • zaneperry, on 04/02/2009, -0/+237They have come a long way.
    http://backrub.c63.be/May1998/hardware.htm

    I actually saw these when I was working on a project in Chicago (peeked down a fenced in Aisle of a Data Center).
  • sublimeparanoia, on 04/01/2009, -4/+197that's actually pretty interesting
  • megamod, on 04/02/2009, -32/+172hmmm...written april 1st.
  • twtmc, on 04/03/2009, -3/+140What part seems like a joke to you?
  • tattokris, on 04/02/2009, -1/+96That made the server in my closet cry.
  • Elementix, on 04/03/2009, -17/+86ugh... this Diggbar ***** is pissing me off
  • inactive, on 04/03/2009, -0/+50Actually I'd love a ***** job like that, in fact I'd love any ***** job right now!
  • inactive, on 04/03/2009, -1/+46mmmm chocolate teapot
  • flammablewater, on 04/03/2009, -0/+44"If Google uses them, they're good enough for your dumb ass" - buy Gigabyte
  • thetanman, on 04/02/2009, -0/+42Any techies know how the hard drives might be configured? RAID striped for speed, cloned for redundancy or just more capacity cheaper?
  • azbmr, on 04/03/2009, -0/+41I wouldn't expect much striping. Google works in a lot of redundancy. They have to. Look at all the "cheap" hardware and take into account that 3% of all hard drives fail in the first three years. The volume of drives alone ensure routine failures. Their custom file system is set up so that data exists in a minimum of three places in the "crate."
  • linagee, on 04/03/2009, -0/+38Velcro - the reusable duct tape.
  • wpyh, on 04/03/2009, -3/+40The 99.9% efficiency?
  • harazek, on 04/03/2009, -0/+36My Profile --> Settings tab --> Viewing Preferences --> Show diggbar?
  • legoalert33, on 04/03/2009, -2/+36Dugg for Lego case.
  • linagee, on 04/03/2009, -0/+34deathstar?
  • jander99, on 04/02/2009, -1/+34They have two hard drives on each server. However, those hard drives are using googleFS. Think of a Raid5 SAN over 1160 servers. Redundancy is builtin to the filesystem. One drive fails, the information is available on another drive somewhere else in the datacenter, or halfway around the world.
  • abrasion, on 04/03/2009, -0/+32I would love to know more about the software side of things.

    I've heard the google FS is all about redundancy, also I believe it's a custom motherboard, what firmware is on it, what special features?
    Does it report drive failures or battery capacity dropping?

    How much does it cost to make each one? Are those drives RAID 0, RAID 1, RAID 5 or something entirely different?
    Just how good is the google"FS" - do you simply add a machine with the 'google OS' on it, attach it to the google network and then other nodes start filling it with data?

    As a tech scavenger I'd love to know what they do with all the hardware, there's a perfectly good CPU and some ram on there, plus some HDD's - I've heard when any component dies, they replace the entire server, it works out cheaper.

    Where is the cupboard with 10,000 perfectly good 60 / 80 / 120gb (?) hard disks attached to dead gigabyte boards?!
    Where's the same cupboard but with 80,000 ram sticks in it.. oh my.

    You could move so much stuff on ebay, it's awesome.
  • ttamshadbolt, on 04/01/2009, -6/+38I'm surprised they used direct attached storage
  • Krumm, on 04/03/2009, -0/+30It will be on trickle charge to keep it topped off, and only switched into the circuit on power failure.
    The key parts here are the custom 12v only PSU (note the google P/N) and the Gigabyte board that has a switching PSU on board (I want one!).
    If you peer close enough, even the drives are powered from the motherboard, not direct from the PSU.

    Very nice design, there must be a monitoring / feedback system to keep track of the battery so they know when to replace them, unless they just bin the systems every few years.
    It would be great to get hold of one of these motherboards, you could run a few from a single PSU for better efficiency, or have a kick ass car system!
  • kopas, on 04/03/2009, -0/+26Video Tour: http://www.youtube.com/watch?v=bs3Et540-_s
  • darkstar949, on 04/02/2009, -5/+30Direct attached storage is still going to be faster than network storage as SATA II has a speed of 3Gb/s where as the fastest network card speeds are currently only 1 Gb/s.
  • inactive, on 04/03/2009, -1/+26nice ad for gigabyte there.
  • mahnic, on 04/03/2009, -0/+25I love how velcro is holding in place the PSU, HDDs and the battery.
  • ShnowDoggie, on 04/03/2009, -1/+2410GB probably cost to much. FTFA:

    ''Early on, there was an emphasis on the dollar per (search) query," Hoelzle said. "We were forced to focus. Revenue per query is very low."

    Mainstream servers with x86 processors were the only option, he added. "Ten years ago...it was clear the only way to make (search) work as free product was to run on relatively cheap hardware. You can't run it on a mainframe. The margins just don't work out," he said. "
  • cshard, on 04/03/2009, -1/+24I have to admit, I did have to do a double-take because of the date. But this seems to be for real.
  • garrettg84, on 04/03/2009, -1/+23@jektal
    64-bit is pretty generic, and absolutely meaningless. if you were referring to x86-64, it is still x86. if you were referring to something else, you have ineffectively stated it.
  • dalectrics, on 04/03/2009, -0/+21Fighting Grammar Nazi urges...
  • HonoredMule, on 04/03/2009, -0/+21Would you rather proudly display your servers in the living room?

    ò_Ó
  • mikedeezy33, on 04/03/2009, -0/+20likely they will replace the entire servers, as the processors and other components will be too outdated. Google is smarter, they win again.
  • angrykeyboarder, on 04/02/2009, -0/+19What processors would you have guessed they used in place of x86-based ones?

  • Suricou, on 04/03/2009, -0/+19Don't last long?

    That's a SLA battery. And it's being trickle-charged, not deep-cycled. Rarely drained at all.

    It could last twenty years or more under that usage. It'll certinly outlast the rest of the server.
  • blue98camaro, on 04/03/2009, -2/+20Way to break the site with the diggbar...
  • Jektal, on 04/02/2009, -6/+24I remember hearing something about Google using shipping-container based data-centers several years ago. I think back then it was in connection to all the dark fiber they were buying, and the idea that they could just deliver a container to a spot, plug it in, and boom: instant local ISP.

    Also surprising that they still use x86 processors and have 8 memory modules per server. Must have gotten a really good price on 512MB sticks.
  • BenKenobi88, on 04/03/2009, -4/+21I dunno, the part where everything's haphazardly velcro'd together?
  • jojo1224, on 04/03/2009, -0/+17The drive's used are Hitachi Deskstar 7K1000 1TB.
  • TVarmy, on 04/03/2009, -0/+16Working in a nice, clean data center, just unplugging old batteries and plugging new ones in? I'd take it. If they let me listen to audiobooks at the same time on an MP3 player, it's a great job. I'm assuming I'd also get the benefits and free lunches that Google is famous for giving employees.
  • mrnathan, on 04/03/2009, -0/+16Well they claim it was a secret, but if Google has a patent (like the article says) on the built in batteries then its public domain. Feel free to go check it out yourself with the patents office.
  • GraceHead, on 04/03/2009, -0/+15is it coming out of the closet?
  • MattCruikshank, on 04/03/2009, -0/+15Why?

    You've got a million computers. They each have a cheap-ass hard drive.

    Do the math - that's an enormous amount of storage capacity, and if you chunk up the data, an unbelievable amount of bandwidth, if you're moving data from machines M to machines N. Delivering the same effective bandwidth to machines N from a non-attached storage would be ridiculously expensive.
  • YellowDucati, on 04/02/2009, -2/+17Geek Mecca
  • metalstorm, on 04/03/2009, -0/+15Too bad they are as thick as my finger, have sensitive jacks, have a large minimum bend radius and the switches for them are HUGE.

    Note: I built a couple of large scaled Inifiband clusters that involved miles and miles of it and it left me bitter.
  • SteveMax, on 04/03/2009, -0/+15AMD/Intel's 64-bit processors (besides the Itanic) are still x86. Of course it's a slightly modified architecture (x86_64), but nowadays it's implied that if you get a x86 processor, it will be x86_64 capable. Google upgrades their servers quickly enough (they're on their 6th or 7th generation since 2005, according to TFA) to be able to use newer technologies. Besides, nowadays it's basically impossible to find new x86 processors that won't do x86_64.
  • inactive, on 04/03/2009, -4/+18What is this? 2007 Digg??? Where am I?
  • dtfinch, on 04/03/2009, -0/+13If the GA-9IVDP is a custom for Google version of the GA-9IVDT, it probably supports 16gb ram and the only processor it'll take is a Xeon with 64-bit support.

    http://www.gigabyte.com.tw/Products/Networking/Pro ...

    And even if it was running at 32 bit, Linux can address 64gb using PAE, with the only limit being 3gb per process.
  • mr5150, on 04/03/2009, -0/+13in breaking news: it has been discovered who is the no.1 ebay seller in the world, he goes by the name of "heaps of ***** to sell" and has just traded his 1 millionth item.....more after the break.
  • itdood, on 04/03/2009, -1/+14Google isn't just a revolutionary search algorithm, they are also doing some bleeding edge stuff with distributed computing and mass storage that most CIOs would be too scared to try. If they decided to use a traditional SAN their costs would probably be around $20m or so per year just for their web index and gmail accounts storage, and they would still need the server clusters... (I read somewhere between those two, it's about 2,200 TB).

    CIOs should be watching this....are we ready to dump EMC yet??
  • proliance, on 04/03/2009, -0/+13Replacing the batteries in a standard size UPS sucks. I'd rather replace small ones like these.
  • cyssero, on 04/18/2009, -1/+14HDMI has a greater throughput than both SATA and Gigabit, but the problems are the cable runs you can achieve. After about 10m on cheap cable you will get loss and it deteriorates rapidly thereafter.

    Cat5/6 runs can go for 100 meters quite easily, so that's why SATA/HDMI isn't going to replace fast ethernet.
  • Fetching more discussions...
    Show 51 - 100 of 264 discussions

  • Add a Comment

    Login or register to comment!
    Or, sign-in super fast with your Facebook account ...


© Digg Inc. 2009 — Content created and posted by Digg users is dedicated to the public domain | Terms of Use Updated | Privacy Policy
DIGG, DIGG IT, DUGG, DIGG THIS, Digg graphics, logos, designs, page headers, button icons, scripts, and other service names are the trademarks of Digg Inc.