[Randtest-develop] Randtest (fwd)

[Mike R. <er...@es...>]

I must have messed up something creating the randtest project, I hope
everyone on the list gets this.  I didn't, so I'm forwarding it.

---------- Forwarded message ----------
Date: Tue, 16 Oct 2001 20:02:53 -0400
From: "John E. Hadstate" <jha...@be...>
To: "Rosing, Mike" <er...@es...>
Subject: Randtest

Mike,

I got a rejection notice on this message when I sent
it this morning.  If you already received it, just
disregard.  Thanks, -jeh


I tend to agree with your feeling that we should
pull the file into RAM for processing.  It will make
the analysis faster up to the point when you run out
of physical memory and start page faulting.  At that
point, it will become _slower_ than the current
implementation because page faulting requires a
write to the page file followed by a read from
someplace else in the page file.  The current
implementation only reads sequentially forward,
hence, the O/S can do lookahead optimizations on the
disk access.

I can't quite make up my mind whether bitwise
statistics are the same as numerical statistics.
What keeps nagging at me is that when a collection
of bits becomes a number, the j-th bit in the number
is weighted by 2^j.  It seems, therefore, that the
statistics of the j-th bit are 2^j times as
important in how the number changes.

I posted the first two mail messages about Randtest
to the SourceForge discussion board and everything
worked swimmingly.

I guess I would like to implement the NIST tests
next.  If you know a URL where I can find the specs,
let me know.  I would rather implement from the
specs than copy somebodies code unless I know that
code is trustworthy.

Regards,

-jeh