PRISM: NSA/FBI Internet data mining project

Matthew Petach mpetach at
Fri Jun 7 16:32:53 UTC 2013

On Thu, Jun 6, 2013 at 5:04 PM, Matthew Petach <mpetach at>wrote:

> On Thu, Jun 6, 2013 at 4:35 PM, Jay Ashworth <jra at> wrote:
>> Has fingers directly in servers of top Internet content companies,
>> dates to 2007.  Happily, none of the companies listed are transport
>> networks:
>> Cheers,
>> -- jra
>> --
>> Jay R. Ashworth                  Baylink
>> jra at
>> Designer                     The Things I Think                       RFC
>> 2100
>> Ashworth & Associates         2000 Land
>> Rover DII
>> St Petersburg FL USA               #natog                      +1 727
>> 647 1274
> I've always just assumed that if it's in electronic form,
> someone else is either reading it now, has already read
> it, or will read it as soon as I walk away from the screen.
> Much less stress in life that way.  ^_^
> Matt

When I posted this yesterday, I was speaking somewhat
tongue-in-cheek, because we hadn't yet made a formal
statement to the press.  Now that we've made our official
reply, I can echo it, and note that whatever fluffed up
powerpoint was passed around to the washington post,
it does not reflect reality.  There are no optical taps in
our datacenters funneling information out, there are no
sooper-seekret backdoors in the software that funnel
information to the government.  As our formal reply
stated: "Yahoo does not provide the government with
direct access to its servers, systems, or network."
I believe the other major players supposedly listed
in the document have released similar statements,
all indicating a similar lack of super-cheap government
listening capabilities.

Speaking just for myself, and if you quote me on this
as speaking on anyone else's behalf, you're a complete
fool, if the government was able to build infrastructure
that could listen to all the traffic from a major provider
for a fraction of what it costs them to handle that traffic
in the first place, I'd be truly amazed--and I'd probably
wonder why the company didn't outsource their infrastruture
to the government, if they can build and run it so much
more cheaply than the commercial providers.  ;P
7 companies were listed; if we assume the
burden was split roughly evenly between them, that's
20M/7, about $2.85M per company per year to tap in,
or about $238,000/month per company listed, to
supposedly snoop on hundreds of gigs per second
of data.  Two ways to handle it: tap in, and funnel
copies of all traffic back to distant monitoring posts,
or have local servers digesting and filtering, just
extracting the few nuggets they want, and sending
just those back.

Let's take the first case; doing optical taps, or other
form of direct traffic mirroring, carrying it untouched
offsite to process; that's going to mean the ability to
siphon off hundreds of Gbps per datacenter and carry
it offsite for $238k/month; let's figure a major player
has data split across at least 3 datacenters, so about
$75K/month per datacenter to carry say 300Gbps of
traffic.  It's pretty clearly going to have to be DWDM
on dark fiber at that traffic volume; most recent
quotes I've seen for dark fiber put it at $325/mile
for already-laid-in-ground (new builds are considerably
more, of course).  If we figure the three datacenters
are split around just the US, on average you're going
to need to run about 1500 miles to reach their central
listening post; that's $49K/month just to carry the
bitstream, which leaves you just about $25K/month
to run the servers to digest that data; at 5c/kwhr, a
typical server pulling 300 watts is gonna cost you $11/month
to run; let's assume each server can process 2Gbps of
traffic, constantly; 150 servers for the stream of 300Gbps
means we're down to $22K for the rest of our support
costs; figure two sysadmins getting paid $10k/month
to run the servers (120k annual salary), and you've got
just $2k for G&A overhead.

That's a heck of an efficient operation they'd have to be
running to listen in on all the traffic for the supposed
budget number claimed.

I'm late for work; I'll follow up with a runthrough of the
other model, doing on-site digestion and processing
later, but I think you can see the point--it's not realistic
to think they can handle the volumes of data being
claimed at the price numbers listed.  If they could,
the major providers would already be doing it for
much cheaper than they are today.  I mean, the
Utah datacenter they're building is costing them
$2B to build; does anyone really think if they're
overpaying that much for datacenter space, they
could really snoop on provider traffic for only

More later--and remember, this is purely my own
rampant speculation, I'm not speaking for anyone,
on behalf of anyone, or even remotely authorized
or acknowledged by any entity on this rambling,
so please don't go quoting this anywhere else,
it'll make you look foolish, and probably get me
in trouble anyhow.  :(


More information about the NANOG mailing list