next up previous contents
Next: Simulation on multi-core computers Up: A Discrete EVent system Previous: Continuous Models   Contents

The Simulator Class

The Simulator class has four functions: to determining the model's time of next event, to extract output from the model, to inject input into the model, and to advance the simulation clock. The first function is implemented by the nextEventTime method with which we are already familiar. I'll address the other three functions in turn.

There are two essential steps for extracting output from your model. The first step is to register an EventListener with the simulator. This is done by creating a subclass of the EventListener and then passing this object to the Simulator's addEventListener method. When the EventListener is registered with the simulator, its outputEvent method intercepts output originating from Atomic and Network models.

The second step is to invoke the Simulator's computeNextOutput method, which performs the output calculations and provides the results to registered EventListeners. The signature of computeNextOutput is

void computeNextOutput()
and it computes the model output at the time given by the nextEventTime method. The computeNextOutput method invokes the output_func method of every imminent Atomic model, maps outputs to inputs by calling the route method of Network models, and calls the outputEvent method of every EventListener registered with the Simulator. The computeNextOutput method anticipates the output of your model from its current state assuming that no input events will intervene between now and the time returned by nextEventTime.

The computeNextState method is used to inject events into a model and advance the simulation clock. The method signature is

void computeNextState(Bag<Event<X> >& input, double t)
where the Event class is the same one that the EventListener accepts to its outputEvent method. The Event class has two fields: a pointer to a model of type Devs$ <$X$ >$ (i.e., a Network or Atomic model) and a value of type X.

The computeNextState method applies a bag of Event objects to the model at time t. If the input bag is empty and t is equal to the next event time, then this method has the same effect as execNextEvent: it calculates the output values at time t using the computeNextOutput method, computes the next state of all models undergoing internal and external events, computes structure changes, and advances the simulation clock.

If the input bag is not empty then the value of each Event is applied as an input to the model pointed to by that Event. If, in this case, t is equal to the next event time then the method also follows the usual steps of invoking the computeNextOutput method and calculating state and structure changes. However, if t is less than the Simulator's next event time, then the procedure is nearly identical excepting that the computeNextOutput method is not invoked. In this case, the only input events for any model are those provided in the input bag.

The Simulator's execNextEvent method does its job using computeNextOutput and computeNextState. The implementation of execNextEvent has only two lines; the Bag bogus_input is empty.

void execNextEvent() {

The computeNextOutput, computeNextState, and execNextEvent methods throw an exception if a model violates either of two constraints: i) the time advance is negative and ii) the coupling constraints described in section [*] and illustrated in Figure [*] are violated. The Adevs exception class is derived from the standard C++ exception class. Its method what returns a string that describes the exception condition and the method who returns a pointer to the model that caused the exception.

The Adevs exception class is intended to assist with debugging simulations. There isn't much you can do at run-time to fix a time advance method or reorganize a model's structure (or fix the structure change logic), but the simulator tries to identify problems before they become obscure and difficult to find bugs.

next up previous contents
Next: Simulation on multi-core computers Up: A Discrete EVent system Previous: Continuous Models   Contents
James J. Nutaro 2017-03-07