Re: advice sought on key/data histogram analysis of rijndael/128 and serpent

okay, i reduced the number of blocks being chained
together to just two, so that means doing a p-value on
256 bits. it also means having to increase the outer
loop quite a bit.

http://lkcl/net/crypto/key_as_data.1.tgz

i just ran a test generating 128x128x102400 p-values,
and analysed the results, a graph of which can be seen
here:

http://lkcl.net/crypto/data.1.rjd.2.204800.histogram.256x32x4.jpeg

using gnuplot scatter.data.dem.

remember, in the plots, i've turned the histogram on its side.
so the x-value is the histogram value (and therefore gnuplot
sorts the data by histogram size along the x-axis), the y-value
is the bit being altered in the data (i think) and the z-value
is the bit being initially set in the key (i think).

i would expect there to be at least _some_ clear randomness
even in the low end of the histogram values: what i'm very
concerned about is seeing clearly distinct curves which very
bluntly show a correlation between the key bits and the data bits.

and, more worryingly, these correlations can be seen not just
at the low and high end of the (sorted) histogram values, but
curves can quite clearly be seen in the entire first and last
THIRDS of the histogram range.

only when you get into the middle range of the histogram
values is it no longer possible to see clearly whether
there's any correlations.

so, what are the options, here? what have i done wrong?
Received on Mon Oct 17 20:48:40 2005