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

[Daniel S. <da...@Op...>]

Peter Wickersham wrote:
> It would appear that the random function used within OpenSTA is not
> very random.
> 
> I was having some issues with using RANDOM on a range of 1 - 120000.

Have you seen/found this bug in the bug DB:
  http://sourceforge.net/tracker/?group_id=10857&atid=110857&func=detail&aid=415144
It looks like you'll never get anything over 32768 anyway.

> It didn't appear to me that it was very random since I seemed to
> get repeating numbers pretty quickly. 

I seem to remember someone talking about this behavior on the Users
list or somewhere a long time ago ... can't find it though and
there's no bug logged.

> So, I looked at the source and the code appears to use srand()
> everytime it calls rand() with the last result from calling rand().
> However, this actually ends up producing more collisions more
> quickly than just calling srand() once. 

Can you give us a source file and line number to save me a little bit
of work?  I'm only just getting into the inner workings of SCL - the
previous SCL internals expert no longer has time to give to this 
project.

> I took the same approach in a sample app and ran a loop for 1000
> times generating random numbers between 1 and 30000.  Using srand()
> once produced 974 unique numbers.  Using srand() every time with
> the last result of rand() produce 173 unique numbers.

Sounds pretty conculsive - the problem is that changing this behavior
means that the REPEATABLE RANDOMs would no longer be ... repeatable,
and this obviously isn't a good thing for a potential users existing
regression test.  We can get around this by making the REPEATABLE
RANDOMs keep the old behavior unless a config file is updated
somehow, so it isn't such a big deal but it does need baring in mind.

> I have not had the opportunity to configure the source to build it
> and test any changes.  In fact, I have the CVS HEAD, so that might
> be problematic if I read the site correctly.  Not to mention that I
> currently have Visual Studio 2003

I'm currently having success using a version built with the current
Platform SDK and no STLport (contemplating dropping my changes into
CVS) but I am using VS6sp6 and have never tried using VS2003 (don't
have it).  If you want to suggest specific code changes then I can
do a test version to make sure this gets sorted.  Please feel free to
log a bug for this on SF.

Cheers
/dan
-- 
Daniel Sutcliffe <Da...@Op...>
  OpenSTA part-time caretaker - http://OpenSTA.org/