[erlang-questions] Implementation of a 2006 version of Wichmann-Hill random number generator for Erlang/OTP

[Kenji Rikitake]

Robert and all:

The reason why I picked up Wichmann-Hill's 2006 algorithm was exactly
what you wrote below.  Here's some observations:

* generated random number range: same (0.0 =< Result < 1.0)

* performance: ~9% slower when running in an infinite loop
  (old: ~67000, new ~60000 samples/sec 
   on Erlang/OTP R14B FreeBSD/i386 Core2Duo 1.8GHz with no HiPE)

* seed elements: old 3, new 4, as Robert pointed out.

  I don't have an idea about generating two elements from one.

  Also I think the seeding function should be

  seed(A1, A2, A3, A4) ->
    put(random_wh06_seed,
	{abs(A1 - 1) rem 2147483578 + 1,
	 abs(A2 - 1) rem 2147483542 + 1,
	 abs(A3 - 1) rem 2147483422 + 1,
	 abs(A4 - 1) rem 2147483122 + 1}).
  
  so that {0, 0, 0, 0} -> {1, 1, 1, 1}
  while {1, 2, 3, 4} -> {1, 2, 3, 4}.
  (The zero values must be avoided)
  
* seed values: old unsigned 16bit, new unsigned 32bit
  (This will not cause any serious issue)

* Licensing issues: the algorithm is openly published in an academic
  paper, though the NPL (the copyright holder) code is not in the open
  license.  I think changing the default seeding (at seed/0) values from
  the NPL code (already done) will be sufficient to avoid the licensing
  issues.

Any other points?

Kenji Rikitake

In the message <832946219.146801291157435038.JavaMail.root@REDACTED>
dated Tue, Nov 30, 2010 at 10:50:11PM +0000,
Robert Virding <robert.virding@REDACTED> writes:
> What would be nice is if we could just drop the new algorithm into the old module so we avoid having to have 2 random modules. As I see it the main problem is that the old random has 3 seeds while the new has 4, which makes the seed functions incompatible. Couldn't you just have one seed be two?
> 
> It would mean that the "new" random would return better values, but they would be different from what the "old" random returned. I don't see this as a problem but others might.
> 
> Robert

> > > > The 2006 Wichmann-Hill RNG for Erlang moved to
> > > > https://gist.github.com/713144
> > > > under the name of
> > > > random_wh06.erl