[Repast-developer] Re: [Repast-interest] BaseController, Controller and BatchController (and Jelly)

[Nick C. <nic...@ve...>]

Hi,

The new IController interface defines methods for executing (starting,
stopping etc.) a simulation. The BaseController implements this
Interface in an asynchronous way, that is, it creates a thread and
executes scheduled actions in this thread. Start, stop etc. merely
(oversimplifying a bit) start and stop this thread. The batch controller
and gui controller add parameter handling on top of this. A new
controller, Thin Controller, adds a bit more that is necessary for a
non-abstract controller, but it still relies on BaseController to set up
the thread. 

So, if you want a controller that not so asynchronous then what repast
supplies won't work. However, it would be fairly easy to construct such
a thing using BaseController as a model. In fact it sounds like all you
want is to create a Schedule add actions to it for execution and then
iterate over the Schedule calling execute some number of times. Again,
looking at the code in Schedule and ScheduleBase would might help you
out here. 

I've often thought that it should be possible to create composite nested
simulations out of pieces that repast provides, and it would be good to
see someone try!

As for posting post-doc announcements etc. repast-interest is a fine
place for that.

Nick

On Sun, 2002-12-15 at 15:23, Andy Cleary wrote:
> [My take on this seemed more appropriate to repast-developer than 
> repast-interest, so I've cross posted it...]
> 
> At 01:35 PM 12/8/2002 -0500, Mark R. Diggory wrote:
> >Hi all,
> >
> >I'm trying to do some development that involves using the Jelly/Ant 
> >scripting env (Apache Jakarta Commons) to build and execute my simulation 
> >in batch processes. Where before I was just execing my model as a Java 
> >Runnable using Ant's Java Task, now I am actaully trying to instantiate 
> >the model class and controller from within Jelly Tags. (Analogous to "JSP 
> >tags in JSP/Servlets" or "Ant tasks in Ant".  Jelly is a really powerfull 
> >XML based scripting environment that can be extended with other scripting 
> >languages (java,javascript,jpython...).
> >
> >http://jakarta.apache.org/commons/sandbox/jelly/
> >
> >I'm finding that the Controller class hierarchy is a little hard to work 
> >with as an Interface/Extensible group of classes.
> >
> >BaseController --> Controller
> >
> >and
> >
> >BaseController --> BatchController
> 
> I know this is a bit late since Nick has already done his refactoring, but 
> I didn't realize until now that this discussion is close to one of my own 
> interests, and I wanted to express that interest and see if anyone has a 
> quick answer as to whether the new refactoring works for this.
> 
> We have an interest in constructing simulations in which many Repast 
> simulations will be parts, but not all, of the simulation. At its simplest, 
> imagine a piece of code sitting on top of a bunch of Repast simulations, 
> acting as the master that tells the individual simulations when it is their 
> turn to execute, for how long, that sort of thing.
> 
> Conceptually, this appears to clearly be a job of a "Controller", in the 
> sense of a well-defined interface, that is, the "master" would want to have 
> access to a "Controller" for each simulation. But there are some 
> differences, I think, between the main two imagined uses of controllers so 
> far. This isn't a batch controller exploring a parameter space that wants 
> to run a simulation all the way to the end and then start up a new one. Nor 
> am I going to be controlling the simulations "asynchronously" via user 
> intervention in a GUI. I want to switch back and forth between simulations 
> with very tight control of how long they simulate.
> 
> I'm still looking at the API documentation that I'm sure existed before 
> this refactoring but it wasn't obvious to me that this would be easy at 
> least given the current documentation. The BaseController functionality 
> seems to be based on a notion that the controller and the simulation itself 
> are pretty asynchronous (the simulation being a different thread), but I 
> wanted tighter control than that, for example to tell a simulation to run X 
> ticks and then return control to the controller.
> 
> In your opinion, will this fit fairly naturally into the refactoring of the 
> controller hierarchy, and by what suggested mechanism?
> 
> I'm interested at a conceptual level whether you Repast developers see a 
> distinction, which is driving this, in the following: we are interested in 
> a "recursion" in which agents themselves may be entire agent-based 
> simulations. I contrast this with what *appears* to me to be the recursion 
> generally allowed for in Swarm/Repast, in which an agent may in fact be a 
> "collection of agents" (correction: I went ahead and looked at Swarm, and 
> they at least discuss this more complicated case in their 
> documentation...). In my opinion, the latter is a "much simpler" subset 
> (perhaps) of the former. In the latter, essentially every agent in the 
> collective agent is merely "doing the same thing". In the former notion, 
> however, you actually need an entire "sub simulation" inside of your agent, 
> the internal agents interact, you need an internal schedule, etc.
> 
> 
> 
> >Option 2:
> >
> >In which case something like this might be better:
> >
> >*Interface Controller*
> >         public void startSim()
> >         public void pauseSim()
> >         public void stopSim()
> >         public void pauseSim()
> >         public void exitSim()
> >         public void get/setModel()
> >         public void get/setSchedule()
> >         public void addSimEventListener(SimEventListener l)
> >         public void removeSimEventListener(SimEventListener l)
> >
> >*BasicController implements Controller*
> >
> >The implementaion of the basic functionality to control the sim. This can 
> >be used to control the sim programatically.
> 
> Right, this is what I am looking for: to control the sim programatically.
> 
> Cheers,
> Andy
> ============================================
> Andrew J. Cleary, Software Engineer/Computational Scientist
> Lawrence Livermore National Labs, 925-424-5890.
> ============================================
> 
> 
> 
> -------------------------------------------------------
> This sf.net email is sponsored by:
> With Great Power, Comes Great Responsibility 
> Learn to use your power at OSDN's High Performance Computing Channel
> http://hpc.devchannel.org/
> _______________________________________________
> Repast-interest mailing list
> Rep...@li...
> https://lists.sourceforge.net/lists/listinfo/repast-interest
-- 
Nick Collier 
Social Science Research Computing 
University of Chicago 
http://repast.sourceforge.net