Re: parallelism

[glen e. p. ropella]

At 10:14 AM 3/28/00 +0200, you wrote:
> Marcus:
> > The goal of high-level parallelism for Swarm requires tools for
> > quantitatively determining which sets of agents are
> > logically/effectively independent and which are interdependent.
>
> Um, wouldn't a [ActionGroup setDefaultOrder: Randomized] suffice
> to tell the system which ones are mutually independent?

Not necessarily.  It certainly helps.  But, since by "Randomized",
we really mean "pseudo-Randomized", all you buy here is an interface
to another piece of code, namely, the random library.  You pay for
this by assuming trust in the random library (which I think Swarm
covers quite nicely due to Sven's effort).  In a parallel system,
your purchase would be different.  Rather than buying your
"dependence discovery" with trust in more software, you buy it
with repeatability.  If Swarm were implemented on a parallel system
(meaning that two or more things can actually happen at the same time),
you run into problems if you want to make a model run exactly the
same way it has in the past.  The up-side, of course, is that if
you get the phenomena you want to see over multiple runs, you can
be fairly certain that your concurrently specified actions don't
have a hidden dependency (because they actually have an implementation
level independence).  The down-side is that you can't control their
execution in a fascist way.

glen

--
glen e. p. ropella =><= The front line is everywhere. Hail Eris!
Home: http://forager.swarm.com/~gepr              (505) 424-0448
Work: http://www.swarm.com                        (505) 995-0818


                 ==================================
  Swarm-Modelling is for discussion of Simulation and Modelling techniques
  esp. using Swarm.  For list administration needs (esp. [un]subscribing),
  please send a message to <address@hidden> with "help" in the
  body of the message.
                 ==================================