Re: Generating good quality 128bit random numbers

From: study (study_at_home.com)
Date: 09/18/05 

Date: Sun, 18 Sep 2005 12:44:26 -0700

On Sun, 18 Sep 2005 10:20:46 GMT, Bryan Olson
<fakeaddress@nowhere.org> wrote:

>study wrote:
> > Bryan Olson wrote:
> >>You don't need random at all. To get a universally-unique
> >>128-bit number, start with a long universally-unique string,
> >>then hash it to 128 bits.
> >
> > What I'm going to ask about is probably not the best way, but my
> > question is more of a "would this work?" kind of thing.
>
>All the ways you had noted would work, as far as giving you
>a distinct ID every time.
>
>
> > If all entities that worked with the database used a pre-chosen key
> > identical amongst them, and they were all able to generate a
> > universally-unique 128-bit string, could they encrypt the string with
> > AES and get unique IDs where there is zero chance of a collision?
>
>If I'm reading that right, then yes, but I don't see the point.
>If you start with distinct 128-bit strings, then AES-encrypt them
>all under the same key, then the 128-bit ciphertexts are also
>all distinct.
>
>Why the AES step? If, as you wrote, "they were all able to
>generate a universally-unique 128-bit string", why are you
>not done at that point?
>

You would be. You are correct.

But wouldn't doing it that way (with AES step) give these two things?

1. Database IDs that were quite random.

2. Hard to invert the IDs if you didn't have the common key and were
not privvy to method of universally-unique 128-bit string selection.

Thank you.