The NSA’s new mega-datacenter facility in southern Utah may be a lot less impressive, in terms of size and capacity, than the hype surrounding it.
According to NSA datacenter blueprints and escape diagrams obtained by Forbes, the physical space allotted to storage and servers within the facility’s four separate datacenters is far less than published guesstimates.
The identical, 100,000-sq.-ft datacenters are housed in four separate buildings, surrounded by facilities for administration, backup batteries and generators, and security facilities (including a kennel for intruder-sniffing dogs), according to Forbes‘ Kashmir Hill. In the center of each datacenter hall is a space marked “MR—machine room/data center,” each of which is surrounded by cooling and power equipment.
Subtracting the space devoted to servers, cooling and support systems leaves only 25,000 square feet in each facility for storage hardware—100,000 square feet in all devoted strictly to storage, according to Forbes‘ estimates.
Assuming an average physical size and storage density for the hardware stuffed into that space—which is still larger all by itself than all but the largest commercial datacenters—100,000 square feet could hold about 10,000 racks of storage gear, according to estimates provided to Forbes by Brewster Kahle, digital librarian and founder of the Internet Archive, which has been backing up the Web since 1996 and making it available via the Wayback Machine since 2001.
Assuming each rack takes up about 10 square feet, the amount of storage the NSA has devoted specifically to storage could hold 10,000 racks of hardware, each of which could hold about 1.2 petabytes of data at current data-density rates, for a grand total (grossly approximated) of 12,000 petabytes, or 11.7 exabytes.
According to a blog post by Kahle, who wanted to check the numbers after reading the supposed mission of the new NSA facility, digital records of every call made in the United States for a full year would take up about 272 petabytes of space. The blog also points to a 2010 report from the Global Information Industry Center at the University of California, San Diego, which came up with an estimate of 1.36 exabytes per year for combined landline and cellular voice calls (PDF).
Twelve thousand petabytes is a lot, but not anywhere close to the yottabyte estimated by some news sources. Forbes quotes Internet Systems Consortium founder Paul Vixie’s estimate that the NSA datacenter might have storage capacity of only 3 exabytes, which assumes 13 square feet per rack, more space between racks and a lower data density.
NSA whistleblower William Binney estimated the center’s capacity at more like 5 exabytes, according to an affidavit filed in support of an Electronic Frontier Foundation lawsuit against the NSA.
However, as experts quoted in Forbes story and elsewhere have pointed out, the amount of floor space described in a blueprint is not a good way to estimate total storage capacity of a completed datacenter. Given changes in layout and equipment selection, they may not even be good estimates of how much floor space the NSA will actually devote to storage.
Plus, given the pace of change in both hardware and software, the space required for an exabyte’s worth of hardware will continue to shrink, just as improvements in data compression and deduplication have slashed the amount of data to be stored in the first place.
And, of course, there are the government-efficiency and construction-uncertainty factors. The NSA datacenter isn’t scheduled to be completed and in operation until the end of September, but it’s not clear if all four datacenters will be finished and operating at the same time, if they’ll come online in stages, or if one or more will be held in reserve for later expansion.
The NSA already runs six datacenters. In May, the agency launched construction of a new, datacenter in Ft. Meade, Maryland, though the 70,000-sq.-ft.facility will be only the first phase of a $900 million project.
The HPCC2 facility, however, will consolidate several other unnamed NSA facilities when it is completed in 2016.
So, ultimately, if the NSA does want to store a yottabyte’s worth of eavesdropping data, and its Utah datacenter won’t be quite up to the task when it goes online, the agency has plenty of other fallback sites to handle the capacity, even without waiting for technology to catch up with the agency’s expansive but still secret ambitions for PRISM and its other eavesdropping projects.