Re: [OctDev] random numbers for distributed computations

[Michael Creel]

On Tuesday 12 April 2005 03:22, Paul Kienzle wrote:
> On Apr 11, 2005, at 9:04 AM, Michael Creel wrote:
> >> You could try the parallel Mersenne Twister program (available from
> >> the
> >> main Mersenne Twister site):
> >
> > I have looked at this. This seems to implement your suggestion to use
> > different parameters for the MT on each machine, letting the user
> > specify
> > them. But since the octave-forge MT gets its initial parameter
> > from /dev/urandom, then this is already done.
> >
> > From what I understand from the previous discussion, and having looked
> > at the
> > rand.cc code in octave-forge, it seems to me that sequences will be
> > independent on different nodes. Basically, it seems to me that the
> > dynamic
> > creator method
> > (http://www.math.sci.hiroshima-u.ac.jp/~m-mat/MT/DC/dc.html)
> > has been implemented, more or less, in octave-forge.
>
> Changing the seed as we do only changes where the sequence starts.
> It does not change the sequence itself.
>
> I can't read the article because the PDF is messed up, but from the
> bits I can read it sounds like they are using a different generating
> polynomial for each twister rather than simply changing the seed.
> This will change the sequence.
>
> - Paul

I think the present o-f implementation is "good enough for government 
work" (at least on Linux, I don't know how Windows machines initiate the 
RNG). I haven't done any calculations, but I'm pretty sure that the 
likelihood of encountering an overlapping sequence is infinitesimally small 
for clusters of the size encountered on Earth. Thanks, M.



-------------------------------------------------------------
Octave is freely available under the terms of the GNU GPL.

Octave's home on the web:  http://www.octave.org
How to fund new projects:  http://www.octave.org/funding.html
Subscription information:  http://www.octave.org/archive.html
-------------------------------------------------------------