Right now, at my insistence, in order to avoid yet
another installation and platform portability headache,
Bill's entropy generation is in pure Java. Because pure
Java only has a millisecond timer, despite the
availability of microsecond timers on every platform
Java is ported to, entropy generation by sampling
scheduling jitter takes way too long. Now that we're
internally using multiple vats for lots of stuff (eg,
background calculation of the outlines in eBrowser), we
can't survive this overhead, at least as experienced on
my current 366MHz laptop.
* Perhaps the new machines are fast enough that the
issue can be postponed? Not likely.
* I should probably revive the old EC native code for
getting microsecond time on MSWindows and Linux/Unix.
How about the Mac?
* Could we generate and store initial entropy at E
installation time, and turn the pseudo-random crank on
each launch of an E process?
This bug is honorarily assigned to Bill Frantz. Should
he become a member (hint hint), then, if he's willing,
this bug should actually be assigned to him. Hard
requirement: whatever we do here must be approved by
Bill as safe.
Log in to post a comment.
