4 ms·

Some quick math: With 2TB drives, that means 500m drives spinning in the building, assuming no redundancy, fs overhead, OS data, etc. (but then again, who the

by timdorr 17y ago

Some quick math: With 2TB drives, that means 500m drives spinning in the building, assuming no redundancy, fs overhead, OS data, etc. (but then again, who the fuck has this sort of production capacity? Definitely not WD or Seagate or any of the major players) If you store then akin to a Sun X4540 server, you fit 12 drives per U of space. In a 50U rack, if you take away 2U for networking, that's 4812=576 drives per rack. If we throw back in some redundancy, that's about 1m racks. Each rack takes up about 10sf of space (~19"6', accommodating for clearances), so that's 10m square feet of datacenter space, not accounting for any sort of power, cooling, or logistics space. Throw back in the extras and you're looking at roughly a mile square of datacenter space. Wow. So, we all understand this is bullshit journalism, right?

MichaelApproved 17y ago
To put it in context. A square mile has 27,878,400 square feet in it. The Empire State Building has 2,158,000 square feet of usable office space. But this also assumes they're going to use HD as standard storage devices. They would probably be using a different technology. It's clearly not possible with current HD capacity. They could also be building for the eventual capacity and not going to immediately support that much data. The comments on the article page makes this and other good points.
- jbellis 17y ago
  > They would probably be using a different technology. not even the NSA can get a yottabyte of some new technology manufactured without it being all over the news.
  - MichaelApproved 17y ago
    The data doesn't have to be actively online. Tape technology is one option to reduce the amount of hardware and space needed.
    - bbgm 17y ago
      and that process is fairly routine at most national labs and supercomputing facilities. Not highly available, but not sure these data need to be.
ghshephard 17y ago
http://www.datacenterknowledge.com/archives/2009/10/30/more-on-the-nsa-utah-data-center/ http://www.datacenterknowledge.com/archives/2009/10/30/more-... "There’s been a lot of media coverage in the past week of the National Security Administration’s plans for a large new data center at Camp Williams in Utah (which we first noted back in July). The NSA is expected to invest more than $1 billion in the project, which is said to include more than 1 million square feet of space. "
omouse 17y ago
You're assuming they're using server units not mainframes. It would be interesting to see the difference in data center size between mainframes and racks. edit: would someone like to explain why I'm being down-voted?
- sparky 17y ago
  The figure of merit here is really hard drives/square foot. Mainframes use the same hard drives as rack-mounted or blade servers, and it is unclear how much the computational substrate matters in the context of storage capacity per square or cubic foot. Additionally, your wording seems to imply that you think a mainframe would be significantly more dense. While you may cut down on extra NICs, power supplies, cases, etc. by moving to mainframes, the high degree of redundancy in mainframes in general, and their storage subsystems in particular, may well cancel out this advantage as it pertains to density.
emmett 17y ago
From 2005 to 2007 to 2009, the largest available hard drive grew from 500 GB to 1 TB to 2 TB. This is the normal historical trend for hard drives[1]. Projecting that trend out a mere 10 years in the future, we anticipate 64 TB drives to be available, reducing requirements to 32k racks. 10 more years in the future and we have 2 PB drives, reducing requirements to a mere 1k racks. The trend of course might slow before then, but the point still stands that if you're planning on a scale of a decade or two, those numbers stop seeming impossible and start seeming inevitable. [1] http://en.wikipedia.org/wiki/History_of_hard_disk_drives#1980s_to_present_day http://en.wikipedia.org/wiki/History_of_hard_disk_drives#198...

by timdorr 17y ago

MichaelApproved 17y ago

To put it in context. A square mile has 27,878,400 square feet in it. The Empire State Building has 2,158,000 square feet of usable office space. But this also assumes they're going to use HD as standard storage devices. They would probably be using a different technology. It's clearly not possible with current HD capacity. They could also be building for the eventual capacity and not going to immediately support that much data. The comments on the article page makes this and other good points.

jbellis 17y ago
> They would probably be using a different technology. not even the NSA can get a yottabyte of some new technology manufactured without it being all over the news.
- MichaelApproved 17y ago
  The data doesn't have to be actively online. Tape technology is one option to reduce the amount of hardware and space needed.
  - bbgm 17y ago
    and that process is fairly routine at most national labs and supercomputing facilities. Not highly available, but not sure these data need to be.

ghshephard 17y ago

http://www.datacenterknowledge.com/archives/2009/10/30/more-on-the-nsa-utah-data-center/ http://www.datacenterknowledge.com/archives/2009/10/30/more-... "There’s been a lot of media coverage in the past week of the National Security Administration’s plans for a large new data center at Camp Williams in Utah (which we first noted back in July). The NSA is expected to invest more than $1 billion in the project, which is said to include more than 1 million square feet of space. "

omouse 17y ago

You're assuming they're using server units not mainframes. It would be interesting to see the difference in data center size between mainframes and racks. edit: would someone like to explain why I'm being down-voted?

sparky 17y ago
The figure of merit here is really hard drives/square foot. Mainframes use the same hard drives as rack-mounted or blade servers, and it is unclear how much the computational substrate matters in the context of storage capacity per square or cubic foot. Additionally, your wording seems to imply that you think a mainframe would be significantly more dense. While you may cut down on extra NICs, power supplies, cases, etc. by moving to mainframes, the high degree of redundancy in mainframes in general, and their storage subsystems in particular, may well cancel out this advantage as it pertains to density.

emmett 17y ago

From 2005 to 2007 to 2009, the largest available hard drive grew from 500 GB to 1 TB to 2 TB. This is the normal historical trend for hard drives[1]. Projecting that trend out a mere 10 years in the future, we anticipate 64 TB drives to be available, reducing requirements to 32k racks. 10 more years in the future and we have 2 PB drives, reducing requirements to a mere 1k racks. The trend of course might slow before then, but the point still stands that if you're planning on a scale of a decade or two, those numbers stop seeming impossible and start seeming inevitable. [1] http://en.wikipedia.org/wiki/History_of_hard_disk_drives#1980s_to_present_day http://en.wikipedia.org/wiki/History_of_hard_disk_drives#198...