Subject: Re: System clock resolution and random numbers
To: John C. Hayward <John.C.Hayward@wheaton.edu>
From: Rick Byers <firstname.lastname@example.org>
Date: 04/30/1997 18:36:20
On Wed, 30 Apr 1997, John C. Hayward wrote:
> > I'm working on an application that needs as-random-as-possible values.
> > I've been using the tv_usec from gettimeofday as the seed. My question
> > is, what is the resolution of tv_usec? Is it possible to say that it will
> > only ever be a certain subset of possible values? I assume it's very
> > difficult to predict the exact value at any given time (because subsequent
> > calls return significantly different values).
> > Is there any better way to make a random number? I don't like the idea of
> > having a 32 bit seed because that limits the possible outcomes to 2^32.
> > Has their been any thought into putting a random number generator in the
> > kernel, and are there any patches to do this? I would think this would be
> > nearly perfect since it would be impossible to find the pattern unless you
> > had root/physical access.
> Use random - it has a larger facility than 2^32 (from man 3 random)
> The random() function uses a non-linear additive feedback random number
> generator employing a default table of size 31 long integers to return
> successive pseudo-random numbers in the range from 0 to (2**31)-1.
> The period of this random number generator is very large, approximately
> This should be large enough for _any_ application.
I am using random, but the problem isn't with the generator itself, it's
witht the seed. random() has to be initialized with srandom wich takes a
32 bit seed for an argument. Same seed, same sequence of random numbers.
I dont understand exactly how initstate() works, so I'm affraid to rely on
it since I have no real way of telling how random the data is (besides
doing billions of samples and comparing them). Anyone care to comment on
Rick Byers Internet Access Worldwide
email@example.com System Administrator
Welland, Ontario, Canada (905)714-1400