From: norm@netcom.com (Norman Hardy)
Date: Sat, 9 Jul 94 20:52:33 PDT
To: Jim choate <cypherpunks@toad.com
Subject: Re: (fwd) Re: BSD random() - any good (source included)?
Message-ID: <199407100352.UAA15973@netcom.netcom.com>
MIME-Version: 1.0
Content-Type: text/plain


At 15:20 1994/07/07 -0500, Jim choate wrote:
...
>Does anybody know of a good test for randomness? I would definitely like to 
>know how good computer RNG's are. Post away!

A good RNG must pass all such tests. The idea of just one test is itself
dangerous. It would be a generous person who would collect such tests and
organize them to a common interface. Only then would you begin to have "one
test": the collection of these tests.

I coded a blum filter a few years ago which requires about 16,000 random
bits. I tried several prngs in various libraries, then implemented several
from literature including Knuth. All of these caused the filter to work at
about half efficiency. I could find no bugs in the filter code. Then I
recalled that there was a DES routine available. I used DES to generate the
random bits. The filter then worked close to the theoritical maximum!