Cybersecurity

Perderabo -
try Blum Blum Shub PNRG. With correctly chosen large primes, it will meet almost any cryptographic need. Choosing the primes is not trivial. And you need gmp or some other BIGNUM.

http://www.ciphersbyritter.com/NEWS2/TESTSBBS.HTM

Jim it has to run fast in ksh. I am developing a decimal arithemtic based multipy with carry. As far as I know, this is new ground. If I was coding in C, I would use the mersenne twister in a minute. I have toyed with trying to code it in ksh, but it's tough and I wouldn't trust my implementation until I understand how it works. Maybe next year. Meanwhile I want to get the MWC working. It is almost as fast as linear congruential generator. It really should be able to rid the world of lcg's. They persist only because they are fast and easily coded. Diehard is the first test that it hasn't passed. Diehard needs about 80,000,000 bits of random data for a single test run. This takes about 10 hours to generate, so progress is slow. I'm pretty sure the problem is arising when I convert the raw random number to an output modulus. But it will take a week or so to be sure.

I wonder why not use one of the already available PRNG solutions like prngd, or one of the kernel based PRNGs?

Are they not good enough, or for portablility reasons?

I'll say it again: it has to run fast in ksh. And no, prngd cannot pass diehard. The output from /dev/random and /dev/urandom also flunks diehard for sunos and linux. Passing diehard is tough.

A sample run:

Code:

>>}###@>---- CvccvE-n
Template EEEEEEEE has 2,992,179,271,065,856 possibilities
Template CvccvE-n has 2,389,338,000 possibilities

Vephu}:6     VICTOR echo papa hotel uniform right-brace colon six
Sawtu]!7     SIERRA alpha whiskey tango uniform right-bracket exclamation seven
Yuywud-0     YANKEE uniform yankee whiskey uniform delta hyphen zero
MuwsaQ:4     MIKE uniform whiskey sierra alpha QUEBEC colon four
Yaqka%.7     YANKEE alpha quebec kilo alpha percent period seven

>>}###@>----

Each of those 2,389,338,000 possibilities must be absolutely equiprobable. And the selections chosen must be completely unpredictable. Achieving both goals simultaneously is not a piece of cake. And yes, RNGs are notoriously non-portable. I want to do all of this in a self-contained portable shell script.

the prev question was posted while I was writing mine

I am looking forward to seeing your script(if I may of course).

I'm getting close. And I'll post the script here when it's finished. One of the rng's has passed diehard. 62 more rng's to test....

early output:

Code:

$> ksh swordfish 
/dev/urandom 120318102 1330735299 628978242 763612848
>>}###@>---- CvccvE-n
QarhuC/8
Qoyvi+?9
Fusto_>7
Takmuw?6
Feflu?-3

generation of pronouncable passwords :

Code:

>>}###@>---- cvvcvv
loonua
faerai
hookoo
diewoe
xoowiu