| Posted on 2007-04-05 06:23:54-07 by dtlavelle |
| Problem with acquiring random seeds |
|
I wrote a script using Math::Random::MT::Auto to acquire random seeds from the
internet and place them into a table within a MySQL database.
Here is my simple schema:
CREATE TABLE randseed32 (
randid bigint NOT NULL auto_increment,
randseed bigint default NULL,
KEY randid (randid)
);
Here is the small perl script:
use Math::Random::MT::Auto qw(rand),qw(rn_info), qw(get_seed set_seed get_state set_state);
my @rand_seed = get_seed();
for my $rand_seed_iter (0..$#rand_seed){
my $rand_seed = $rand_seed[$rand_seed_iter];
$dbh->do ("INSERT INTO randseed32 SET randseed = $rand_seed");
}
When I run this script for 100 iterations waiting 10 seconds between each script system call, I obtain a table of 25,000 32-bit integers. The unexpected outcome was that out of these 25,000 there were 3093 entries with a count of 2. It would seem that the web site (http://www.randomnumbers.info/) may recycle numbers or maybe I am completely missing something important.
I made a different database and changed the random seed source to 'random_org' (http://www.random.org/). I obtain a total of 32804 32-bit integers. Of these only 20 are repeated, but each of these are repeated 49 times!
+----------------+------------+
| count(randseed)| randseed |
+----------------+------------+
| 49 | 538979961 |
| 49 | 543515987 |
| 49 | 543516788 |
| 49 | 543584032 |
| 49 | 544370534 |
| 49 | 544437353 |
| 49 | 544567129 |
| 49 | 1633972084 |
| 49 | 1634738273 |
| 49 | 1635020661 |
| 49 | 1646292335 |
| 49 | 1679848047 |
| 49 | 1684955506 |
| 49 | 1702065440 |
| 49 | 1702257000 |
| 49 | 1713399143 |
| 49 | 1767994469 |
| 49 | 1870209124 |
| 49 | 1897951861 |
| 49 | 1953461617 |
+----------------+------------+
20 rows in set (0.00 sec)
This seems to be the result of exhausting the daily limit from random.org which is currently set for 1,000,000 bits. I saw no errors dumped to screen while running the script. However, it would seem that the 49 duplicates of the 20 integers are from seeds using the same numerical inputs (is this from the time,pid of the master script used to call the above script every 10 seconds?)
I do not see repeated numbers if I draw 100 iterations of seeds from /dev/urandom. Am I completely missing something and doing something blatantly wrong? Have others seen this as well?
I now have a couple of questions.
1) Does it make any sense to seed a MT with pseudo random numbers from /dev/urandom?
2) Is a MT pseudorandom number generator better than the standard perl rand() function when using only one 32-bit seed value?
I have a project that I would like to generate about 20 million random numbers and possibly the MT algorithm may be a bit overkill. Any hints or suggestions would be greatly appreciated. Thank you. |
| Direct Responses: 4789 | Write a response |
