Re: Summary of discussion of harvester/random locking andperformance optimization

Robert Watson wrote:
[ ...yay! much stuff about S = k log W... ]
> Finally, a question I raised by e-mail previously and think it's worth
> thinking about is how to decide what impact we're having on entropy
> quality.  In particular, are there scores or other easily understood
> numerical or statistical results that can be exposed to a user or
> developer to allow them to measure the impact on entropy quality and
> availability resulting from specific configuration, optimization, etc.

I think what you're asking about is something like:

http://csrc.nist.gov/rng/

...which have a fairly straightforward set of descriptions on a link off of 
that page, as well as source code and datasets.  The code compiles with 
minimal changes-- use gmake and #ifdef 0 two sections of code involving 
definitions for INFINITY and NAN which already exist in the system header files.

There's something about the way these tests abstract away the results into the 
"distribution of P-values" which is not entirely easy to understand, 
unfortunately, although some experimentation reveals that the stats from using 
data from /dev/random are fairly distiguishable from those of feeding the 
output of random() or rand() (or /usr/share/dict/*) into a file.

-- 
-Chuck