Filling up the storage to be used as source for dmcrypt using the
output from /dev/urandom can take a long time since the latter is
quite slow. openssl rand is a few times faster, but I figured that
I'd like something still faster than that, here's the result.
It intends to be using complementary-multiply-with-carry by George
Marsaglia,
taking code posted on
gamedev.net.
On the latter page there has been discussion of not understanding that
implementation. So, don't rely on it producing unpredictable data. In
an attempt to ensure even distribution even if the used algorithm is
partially faulty, the program is mixing the result of the Marsaglia
generator (or ripoff?) with some output from /dev/urandom (using
bitwise XOR). This may be a moot exercise, though.
At least the resulting program is fast enough for my purpose, it generates about 100MB/s on an Atom D510 @ 1.66GHz.
This has been written and tested on Debian only.
Run
make
time ./fastrandom |head -c 100000000|wc -c
for testing, or
make install
for installation.