RE: "Reference #"

[Buck Calabro <Buck.Calabro@xxxxxxxxxxxx>]

Booth wrote:

>If you want uniqueness just go for a random number.  
>The chances of a random number duplicating are slim, 

This is a subject near and dear to my heart.  Apologies for the bandwidth...
I was once tasked with the mission of generating pseudorandom numbers to
pick employees for random drug testing.  The hard part of the task was that
the employees had to be chosen at random.  Provably so.  So my random number
routine (S/38 days - no C API) had to generate stream of pseudorandom
numbers that was mathematically provable.  So it could hold up in
court-provable.  I used the October 1988 issue of Communications of the ACM,
article on random numbers authored by Stephen K Park and Keith W Miller as
the basis of my routine (which held up in court, as it turned out.)

The point is that "random" in this sense does not mean "no duplicates", but
it means that one cannot predict the next number in the sequence.  It is
VERY likely that you will generate random numbers, depending on the universe
you are choosing from.  Generally, people who know about the topic (not me!)
prefer the term "pseudorandom" because a given algorithm will generate the
same sequence of numbers given the same starting point in the sequence.
With a sequential number, one can readily predict the next number in the
sequence, but one can also be 100% certain there are no duplicates until you
run out of numbers in the universe (i.e. if you choose a two digit number,
you'll run out quickly!)
  --buck