Re: parallelism

[Marcus G. Daniels]

>>>>> "GR" == glen e p ropella <address@hidden> writes:

GR> I suppose the question I'm trying to raise, here, is "do we want
GR> to make the machine/model a close mapping?  or do we want to go
GR> full-on for logical abstraction and build things like virtual
GR> machines?"

Given the current logical concurrency model, I can see there is a
detachment of control that could occur as soon as behind-the-scenes
thread stealing scheduling acts on the pool of threads; it is
out of the modeler's hands at that point; a different set of OS-kernel-level
algorithms determine what goes where.  Control could be kept if the
actions that were scheduled had information in them which stated
migration constraints, or if that information was computed based on 
the Swarm activation hierarchy.  Another idea would be to use a Swarm
model to do the migration/stealing, and maybe have some protocol by which
user models could communicate with it.  In either case, the goal would
be to give the OS high-level information about the structure of the model,
so that migration could be done in a smart way (as opposed to damage control).



                  ==================================
   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.
                  ==================================