RE: [OpenSTA-devel] Re: When is Random not random?

[Wickersham, P. <Pet...@di...>]

The issue as I see it is the following. Let's say we have defined two
SCRIPT variables as REPEATABLE RANDOM. A and B with 2 different seeds A1
and B1

Now, lets say that if you called srand(A1), you would see the following
sequence of numbers from rand(). a2, a3, a4, a5, a6.=20

Now, let's say that you called srand(B1) and would normally see the
following sequence of numbers on successive calls to rand(): b2, b3, b4,
b5

First off, srand() will reset the seed for all calls to rand() for all
threads in the process. So, a sequence like below would occur

<initialize A with seed>
GENERATE A =3D> a2
<initialize B with seed>
GENERATE B =3D> b2
GENERATE A =3D> b3
GENERATE B =3D> b4

OR

<initialize A with seed>
GENERATE A =3D> a2
GENERATE A =3D> a3
<initialize B with seed>
GENERATE B =3D> b2
GENERATE B =3D> b3
GENERATE A =3D> b4

You can see how the actual sequences for A and B could often not repeat
just depending on when one or the other was called.

I would imagine that you would want to always be able to count the
sequence of values generated for A and for B to be the same across
multiple runs. I would like to always see A =3D> a2, a3, a4, a5 and B =
=3D>
b2, b3, b4, b5=20

Now, some *nix systems have a rand_r() which is meant to allow a process
to have multiple sequences seeded at different starting points. But, I
don't see this in Windows.

-peter

PS. Looking at the current code, I am guessing that they had this
constant calling of srand() in order to deal with this issue. However,
the result of rand() is not usually the next seed in the sequence within
the actual system call. So, this workaround doesn't work very well.

-----Original Message-----
From: ope...@li...
[mailto:ope...@li...] On Behalf Of Mark
Elam
Sent: Friday, January 06, 2006 1:14 PM
To: ope...@li...
Subject: Re: [OpenSTA-devel] Re: When is Random not random?

Wickersham, Peter wrote:
>=20
> In fact, there already is an issue with this code for REPEATABLE
RANDOM
> and that is the fact that we are using rand() in the first place. This
> is a global function and the seed is a global, as well. So, if you
want
> more than one REPEATABLE RANDOM sequence, this won't work.

I see what you mean, but if the two lines
srand (_cur_seed);
_cur_seed =3D rand();
were an atomic operation, wouldn't that take care of the problem as far=20
as repeatable random is concerned?

Then the #define VOREPRAND could be checked for in _options as to=20
whether repeatable or otherwise (no srand() call) should be used.

Or have I missed something obvious?

Mark




-------------------------------------------------------
This SF.net email is sponsored by: Splunk Inc. Do you grep through log
files
for problems?  Stop!  Download the new AJAX search engine that makes
searching your log files as easy as surfing the  web.  DOWNLOAD SPLUNK!
http://ads.osdn.com/?ad_id=3D7637&alloc_id=3D16865&op=3Dclick
_______________________________________________
OpenSTA-devel mailing list
Ope...@li...
https://lists.sourceforge.net/lists/listinfo/opensta-devel