31 October 2013
NSA MUSCULAR: What else do we know?
Washington Post images of NSA's WINDSTOP-MUSCULAR-INCENSER:
Date: Wed, 30 Oct 2013 19:29:50 -0700 (PDT)
Subject: More on MUSCULAR
What else do we know about MUSCULAR?
In light of a recent Washington Post article detailing how MUSCULAR (DS-200B)
is being used to collect traffic on Yahoo! and Google datacenters, I have
compiled a few additional facts from previous articles that reference
* It is a SIGAD, operated jointed with the GCHQ. It is located "overseas,"
it has a "distributed data distribution system." The documents I have do
not describe in any further detail.
* The "full take" GCHQ site referenced is TEMPORA.
* SCISSORS, which is aimed at reducing the volume of collection is enabled
on MUSCULAR. In particular, SCISSORS is known to block ownerless address
books. The ownerless addressbook elimination function in SCISSORS was implemented
on MUSCULAR on 29 February 2012
** After the implementation of this function in SCISSORS, it was collecting
addressbooks at a rate that would correspond to about 500,000 per year.
(see attached graph) Before the implementation of SCISSORS, the collection
on a single day (10 January 2012) was 311,113 address books. Which, IF this
is truly a representative day, and not an abnormal spike would correspond
to 113,556,245 in a year. SCISSORS matters a lot.
**60 days at 311,113 per day + 306 days at 1388.21 per day (daily
average over 40 days) = 19,091,573 in 2012 (it was a leap year), in real
terms, this means about 19 million address books from that source in 2012.
The 2013 rates should be a lot lower.
*It does NOT contribute to BOUNDLESSINFORMANT, this has to do with
its architecture being different. Or did not at the time the FAQ was written.
So as absolutely mind boggling as the numbers in BOUNDLESSINFORMANT are,
even this does not fully account for the NSA's collection.
About the Address Book Graph:
This is a rather
messy graph that was published in the Washington Post's story about Address
book collection. There are many SIGADs listed. The methodology as usual is
to measure the points along the graph and work out the appropriate scaling.
Because there are a number of lines that cannot be followed, or in some cases
found, the grand total is a bit short of the true total, the lines that can't
be found or followed tend do be clustered near the bottom, so they don't
throw off the total to a huge degree. Another issue with the chart is that
the colors in the chart do not always align perfectly with the key.
A third issue: because lines often intersect, and may be obscured, the accuracy
is plus or minus three pixels, rather than one. In a few places, the position
of the dots were determined by working out the slope of the line. The graphical
chart I have included is of MUSCULAR only.