[Cryptech Tech] Progress on randomness testing, and upcoming TROOPERS conference

[Benedikt Stockebrand]

Hi Joachim and list,

Joachim Strömbergson <joachim at secworks.se> writes:

> Thanks for status and info. One idea I've played with is if it would be
> possible to use GPUs to accelerate randomness testing.

the "number cruncher" PC I've got before the Troopers actually has a
rather over the top GTX980 in it; apparently it's too new to be
supported pretty much anywhere, but at least in the mid term I consider
that an option to investigate if a hexacore CPU can't cut the mustard.

> There is a preso about using GPU to accelerate testing:
>
> http://opengpu.net/EN/attachments/154_HiPEAC2012_OpenGPU_MassiveRand2.pdf

I'll have a look at it as soon as time permits.

> The preso seems to have missed that Dieharder exists, which makes me
> cautious. But the results seems applicable and interesting. Running
> complete Dieharder takes really long time so acceleration makes sense.
> As long as one can implement the tests.

To my understanding one of the shortcomings of Dieharder is that it runs
all tests in sequence, causing (a) a lot of unnecessary I/O at least on
machines with less RAM than the input size and (b) a lot of CPU
cores/threads sitting there and watching a single one of them doing all
the work.  That was reasonable in the late 90s, but today it's rather
inefficient.

That's another thing I want to address with that framework: The next
step is a Pthread wrapper to deal with large numbers of tests running in
parallel all over the place.

Question right now is when I'll find the time and quiet to get this
done.  Life is still pretty busy here...


Cheers,

    Benedikt

-- 
Benedikt Stockebrand,                   Stepladder IT Training+Consulting
Dipl.-Inform.                           http://www.stepladder-it.com/

          Business Grade IPv6 --- Consulting, Training, Projects

BIVBlog---Benedikt's IT Video Blog: http://www.stepladder-it.com/bivblog/