Path: utzoo!mnetor!tmsoft!torsqnt!news-server.csri.toronto.edu!cs.utexas.edu!uwm.edu!psuvax1!news
From: flee@cs.psu.edu (Felix Lee)
Newsgroups: comp.lang.perl
Subject: Re: random numbers?
Message-ID: <3#3Gz8.j@cs.psu.edu>
Date: 10 Feb 91 16:36:24 GMT
References: <16698@venera.isi.edu>
Sender: news@cs.psu.edu (Usenet)
Organization: Penn State Computer Science
Lines: 50
Nntp-Posting-Host: dictionopolis.cs.psu.edu

>Is there some problem with perl's srand and rand functions?

srand() with no arguments does srand(time), and the current time of
day is not a very good seed.

One problem is it doesn't change quickly enough.  The following:
	#!/bin/csh -f
	repeat 10 perl -e 'srand; printf "%.3f ", rand(10);'
will tend to print things like:
	3.158 3.158 8.296 8.296 8.296 8.296 8.296 8.296 3.435 3.435
All the "8.296"s were produced in the same second.

Another problem is the current time of day is monotonically
increasing.  rand() is a simple linear congruential random number
generator (LCRG), typically something like this:
	seed = (seed * 1103515245 + 1234) & 0x7fffffff;
	return seed;
Perl takes this result and divides it by 2**31, turning it into a
floating point number between 0 and 1.

A simple linear function applied to simple, monotonically
increasing values will produce simple results.  The following:
	#!/usr/bin/perl
	for $s (1..10) {
		srand $s;
		for (1..8) {
			printf "%.3f ", rand(10);
		}
		print "\n";
	}
will print something like:
	5.139 1.757 3.087 5.345 9.476 1.717 7.022 2.264
	0.277 6.963 3.125 3.941 7.885 8.269 9.148 8.504
	5.416 2.169 3.163 2.537 6.293 4.821 1.274 4.743
	0.555 7.375 3.202 1.133 4.702 1.372 3.399 0.983
	5.693 2.581 3.240 9.728 3.111 7.924 5.525 7.223
	0.832 7.787 3.278 8.324 1.519 4.475 7.651 3.462
	5.971 2.993 3.317 6.920 9.928 1.027 9.776 9.702
	1.109 8.199 3.355 5.515 8.336 7.579 1.902 5.941
	6.248 3.404 3.393 4.111 6.745 4.130 4.027 2.181
	1.386 8.610 3.432 2.707 5.154 0.682 6.153 8.421
Note how predictable each column is, especially column 3.

Doing something like:
	srand(time + 251 * $$);
	rand; rand; rand; srand(rand(32767));
makes the problem less apparent.  But LCRGs suffer from many flaws.
Avoid rand() if you want good random numbers.
--
Felix Lee	flee@cs.psu.edu