The point isn't to interleave two (or more) different sequences for the benefit
of randomness. The point is to try find as many bugs as possible. With Gremlins
Classic, there were two problems with trying to do that:
Problem #1: Running a Gremlin and then leaving for the night would often result
in the Gremlin finding a bug and then sitting there for the rest of the night,
idling when it could be off doing other things like running other Gremlins. Even
if it didn't find a bug and ran for 1,000,000 events, it could still be sitting
there for hours until you came back.
Solution #1: Gremlin Hordes, where Poser switches to a new Gremlin after a
previous Gremlin finds a bug or runs to completion.
Problem #2: We've noticed that a Gremlin would often find a bug immediately,
within the first 30,000 events, or not at all. Thus, the Horde mechanism would
benefit (that is, find the most number of bugs) by doing some sort of "breadth
first" search instead of running a single Gremlin to completion first. After
all, if a Gremlin didn't turn up a bug in the first 30,000 events, why spend
time running it for the next 970,000 events if you could just switch to another
Gremlin and increase the chance that you'll find a bug?
Solution #2: The "Switching Depth", where you can specify how deep the "breadth
first" search is.
-- Keith Rollin
-- Palm OS Emulator engineer
Richard Hartman <[EMAIL PROTECTED]> on 12/09/99 07:04:58 PM
Please respond to [EMAIL PROTECTED]
Sent by: Richard Hartman <[EMAIL PROTECTED]>
To: "'palm-dev-forum @3com.com'" <[EMAIL PROTECTED]>
cc: (Keith Rollin/HQ/3Com)
Subject: RE: POSE 3.0a3 - Unreliable ???
I'll agree that you may need to test with more
than just gremlin #0 ... but interleaving two
different sequences every 1000 values? I'm still
not sure I see that that is anything more than
just a different pseudorandom sequence. It seems
as if it is just an extra level of complexity that
does not add value. You can get all sorts of
pseudorandom sequences depending upon the seed
you choose ... is interleaving two different
pseudorandom sequences any "more" random than
one sequence?
--
-Richard M. Hartman
[EMAIL PROTECTED]
186,000 mi./sec ... not just a good idea, it's the LAW!
> -----Original Message-----
> From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]]
> Sent: Thursday, December 09, 1999 10:34 AM
> To: [EMAIL PROTECTED]
> Subject: RE: POSE 3.0a3 - Unreliable ???
>
>
>
>
> Is any pseudo-random number generator random enough? There
> are plenty of
> case-examples where an application will survive 1 million
> events at Gremlin #0,
> but fail on other Gremlins. Here at Palm, we like to do as
> much testing as
> possible.
>
> There's even the (very valid) argument that Gremlins testing
> it not enough when
> it comes to automated testing. That's why the scripting
> facility was added -- so
> that you can ensure that all parts of your application are
> testing in every
> conceivable way for which you feel like writing a test case.
>
> -- Keith Rollin
> -- Palm OS Emulator engineer
>
>
>
>
>
>
> Richard Hartman <[EMAIL PROTECTED]> on 12/09/99 09:25:22 AM
>
> Please respond to [EMAIL PROTECTED]
>
> Sent by: Richard Hartman <[EMAIL PROTECTED]>
>
>
> To: "'palm-dev-forum @3com.com'" <[EMAIL PROTECTED]>
> cc: (Keith Rollin/HQ/3Com)
> Subject: RE: POSE 3.0a3 - Unreliable ???
>
>
>
>
> And what is the point of this? Is the random tapping
> of a single gremlin sequence not random enough?
>
> --
> -Richard M. Hartman
> [EMAIL PROTECTED]
>
> 186,000 mi./sec ... not just a good idea, it's the LAW!
>
>
> > -----Original Message-----
> > From: Frykholm, Niklas [mailto:[EMAIL PROTECTED]]
> > Sent: Wednesday, December 08, 1999 11:43 PM
> > To: '[EMAIL PROTECTED]'
> > Subject: RE: POSE 3.0a3 - Unreliable ???
> >
> >
> > >5) I looked at the count and it was 395,000. Some time
> later it was
> > >400,000. Then a few minutes later it was back to 390,000. Very
> > >irratic.
> >
> > It's not a bug, it's a feature! The latest version of POSE
> > has something
> > called "gremlin hordes", which means it switches between
> > several different
> > gremlins. If the switching depth is set to 10000 (which I
> > believe is the
> > default) POSE runs each Gremlin for 10000 events and then
> > switches to the
> > next.
> >
> > This is probably what happened in your case. POSE ran events
> > 390000-400000
> > of gremlin #0 and then events 390000-400000 of gremlin #1.
> > That is why the
> > counter was seemingly set back. (If you studied the dialog
> > box carefully you
> > would notice that the gremlin number in the upper right
> > corner changed.)
> >
> > // Niklas
> >
>
>
>
>
>
>
