From: (Donald T. Davis)
Newsgroups: sci.crypt

Subject: Re: random numbers from disk drive rotation latency -- help?
Date: 14 Jun 1994 01:39:05 -0400
Organization: OpenVision Technologies, Inc.
Lines: 37
Message-ID: <2tjfpp$>
References: <2t06j7$> <2t33p6$> <2t

Colin Plumb writes:
> Donald T. Davis  wrote:
>> our paper ... explains why this works, and discusses
>> our experimental measurements of the noise we found in a unix
>> workstation's disk-rate.
>Well, Crypto '94 hasn't happened yet, as far as I know, so it'll
>be a while.  Are preprints available?

sorry, but crypto makes authors promise not to publish or submit their
articles elsewhere before the conference; until i hear otherwise,
i'm assuming that disseminating it on the net counts as publication.

>> we got 100 bits/minute, plus or minus 1. which we measured in the
>> raw disk-periods with an entropy estimator.
>I'd like to see your arguments and entropy estimator.  I really
>should learn about entropy estimation to beef up PGP's true random
>number generator with better entropy estimates.  the current
>scheme is lamentably ad-hoc.
for the entropy estimates, i recommend ueli maurer's, which i
cited and summarized in this group a month or two ago. his
paper was in crypto '91. please realize, though, that entropy
estimation isn't any more powerful at detecting structure than
standard statistical measures are. you should do the entropy
estimation with your structured data (keystroke timings, e.g.),
and not with your rng output, because to detect lapses of
randomness via an entropy estimate would take lots of memory
and time. one way to look at it is that statistical measures
look for structure, while entropy estimates look for noise.
if you want to separate the signal from the noise, you should
be very careful not to stir the two together.

                                                -don davis