back to article Taiwanese spill on Zuck's racks: Servers powering Facebook REVEALED

Facebook runs on four different server designs, each created to fulfil a specialist task, according to Mike Yang, a general manager and vice president of Taiwanese OEM Quanta. Quanta makes about a quarter of the world's laptop computers and has, of late, branched out into custom servers for big operators like Facebook. …

COMMENTS

This topic is closed for new posts.
  1. Destroy All Monsters Silver badge
    Coat

    “remove anything in our servers that does not contribute to efficiency.”

    Zuck removal happened immediately afterwards.

  2. Pen-y-gors Silver badge

    “remove anything in our servers that does not contribute to efficiency.”

    the data?

  3. Infidellic_

    Let's speculate

    I'm guessing:

    1) I/O work

    2) CPU work

    3) Lightweight proxies

    4) ?

  4. K
    Joke

    The author attended the summit as a guest of Intel

    which paid for flights and accomodation.

    Oh you also forgot.. Hookers, alcohol, narcotics and get out of jail free card!

    1. handle

      Re: The author attended the summit as a guest of Intel

      I presume they actually paid for accommodation.

      What's happened to the "send corrections" link?

      1. Alistair
        Joke

        Re: The author attended the summit as a guest of Intel

        @handle - someone suggested a correction to remove it?

  5. Mr.Si

    Efficiency

    How about Facebook is removed from the facebook servers to increase efficiency.

    1. squigbobble
      Joke

      Re: Efficiency

      We're talking about the efficiency of fb's backend, not the efficency of office workers.

  6. David Kelly 2

    FCC?

    Clearly the greenies at Facebook are not worried about polluting the electromagnetic spectrum if they don't bother putting covers over their servers.

    1. FrankAlphaXII
      Devil

      Re: FCC?

      (First, disclaimer regarding my background, I'm a Signals Intelligence Collector, I do this for a living two weeks a year and weekend a month. Unfortunately, most of the hard math and examples behind this are classified in my case, so maybe a Civilian radio, microwave, or TV transmission engineer or a HAM who isn't so restricted can explain it in depth. My apologies, but I dont want to join Bradley Manning in Leavenworth).

      I kind of doubt they worry about it in a datacenter that they own. Plus, with the frequencies as high as they are but without much power or a "real" transmitter behind them, they most likely cannot propagate too far in distance or into bands that are in use for communications outside of a small area. If they do, there's a major problem and the FCC needs to correct them.

      Its probably noisy as hell in certain EM bands on their property but I doubt you'd be able to tell about 500 meters to a kilometer away except for maybe some residual noise on UHF (if there's a radio or TV tower nearby, it may be VERY obvious and obnoxious to the operator, but I doubt Facebook's stupid enough to do something like that and even then, conditions would have to be next to perfect to notice there was something interfering, due to the height and power behind those types of transmitters, plus they may have their neighbors filtering, Farcebook has quite a lot of money).

      It might be something for the FCC to look into if it causes any disruption to their neighbors signals, or to other companies if the servers they have in CoLo centers also do not have their covers on, but I doubt it does. Its not something I would call a best practice because its an information and physical security risk ("Nice server you have there without covers on it, but you know what? It looks thirsty. I'm gonna give it a gallon of water, all across its motherboard, how's about that?") but I dont see any harm coming to others besides Facebook itsself and their users from this.

      Still, it makes it somewhat interesting from a theoretical TEMPEST point of view, I wonder what transient signals you could intercept from their servers with them leaving the covers off. Too bad that would be harder than hell to research as a private citizen without risking a prison sentence.

      1. Brian Miller

        Re: FCC?

        Oh, it's easy to research it as a private citizen. That's been done and published. You don't monitor the servers, you monitor the LCD screens. Modern CPUs are running in multiple gigaherz using multiple cores, so the RFI would be worthless for monitoring. Yes, such things could be done in the days of the 6502, 6809, 68000 et al, but once CPUs got out-of-order execution and instruction caching, it was all over. After all, if you can't tell what's going on inside the chip when you have full bus access with a HP data analyzer, don't expect the NSA to do it from a block away.

        And as far as sabotage, compare a saboteur vs knuckle-head developers. Who produces more down time?

      2. Solmyr ibn Wali Barad

        Re: FCC?

        In simpler terms - instead of shielded servers, they opted for a shielded building. Or a room.

        But it leaves a problem of interference between servers.

    2. Brian Miller

      Re: FCC?

      What jurisdiction does the FCC have in Taiwan? None, I bet. And if the servers were in the US, I can assure you that a huge datacenter like that would be irrelevant to the FCC. The machines aren't actually broadcasting anything, and outside of the building a flaky street light would probably generate more RFI than those machines inside a metal building.

  7. John Smith 19 Gold badge
    Go

    So they've implemented the Bell Labs Plan 9 architecture?

    Quite sensible really.

    But the EMI in the room is likely to be fierce.

This topic is closed for new posts.

Biting the hand that feeds IT © 1998–2021