SimPy’s Advanced Object Oriented API

Authors:
SimPy release:

2.2

Python version:

2.3 and later (not 3.0)

Revision:

$Revision: 430 $

Date:

$Date: 2010-04-01 03:31:23 +1300 (Thu, 01 Apr 2010) $

Introduction

This document describes the advanced object oriented (OO) programming interface introduced with SimPy 2.0. This is an add-on to the existing API, an alternative API. There is full backward compatibility: Programs running under SimPy 1.9.1 and earlier releases work unchanged under version 2.0 and later versions.

Motivation

Many simulation languages support a procedural modelling style. Using them, problems are decomposed into procedures (functions, subroutines) and either represented by general components, such as queues, or represented in code with data structures.

There are fundamental problems with using the procedural style of modelling and simulation. Procedures do not correspond to real world components. Instead, they correspond to methods and algorithms. Mapping from the real (problem) world to the model and back is difficult and not obvious, particularly for users expert in the problem domain, but not in computer science. Perhaps the greatest limitation of the procedural style is the lack of model extensibility. The only way in this style to change simulation models is through functional extension. One can add structural functionality but not alter any of its basic processes.

Right from its beginning, SimPy, on the other hand, has supported an object oriented approach to simulation modelling. In SimPy, models can be implemented as collections of autonomous, cooperating objects. These objects are self-sufficient and independent. The actions on these objects are tied to the objects and their attributes. The object-oriented capabilities of Python strongly support this encapsulation.

Why does this matter for simulation models? It helps with the mapping from real-world objects and their activities to modelled objects and activities, and back. This not only reduces the complexity of the models, it also makes for easier validation of models and interpretation of simulation results in real world terms.

The new API allows different, often more concise, cleaner program patterns. It strongly supports the development of libraries of model components for specific real world domains. It also supports the re-use and extension of models when model specifications change. In particular larger SimPy programs written with the advanced OO API should be easier to maintain and extend. Users are advised to familiarize themselves with this programming paradigm by reading the models in the SimPyModels folder. Most of them are provided in two implementations, i.e. in the existing and in the OO API. Similarly, the programs in the Bank tutorials are provided with both APIs.

The advanced OO API has been developed very elegantly by Stefan Scherfke and Ontje Lünsdorf, starting from SimPy 1.9. Thanks, guys, for this great job!

Readers of this document should be familiar with the basics of SimPy and have read at least “Basic SimPy - Manual For First Time Users”. They should also know how subclassing is done in Python.

Basic SimPy OO API Design

A class Simulation has been added to module SimPy.Simulation. SimulationTrace, SimulationStep and SimulationRT are subclasses of Simulation. Multiple instances of these classes can co-exist in a SimPy program.

Backward compatibility

Since SimPy 2.0, the package offers both the old/existing API and an advanced object-oriented API where simulation capabilities are provided by instantiating Simulation. SimulationTrace, SimulationStep or SimulationRT are subclasses of Simulation.

Each SimulationXX instance has its own event list and therefore its own simulation time. A SimulationXX instance can effectively be considered as a simulated, isolated parallel world. Any Process, Resource, Store, Level, Monitor, Tally or SimEvent instance belongs to one and only one world (i.e., Simulationxx instance).

The following program shows what this means for API and program structure:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
from SimPy.Simulation import *
"""Object Oriented SimPy API"""

## Model components -------------------------------

class Car(Process):
    def run(self,res):
        yield request,self,res
        yield hold,self,10
        yield release,self,res
        print "Time: %s"%self.sim.now()

## Model and Experiment ---------------------------

s=Simulation()
s.initialize()
r = Resource(capacity=5,sim=s)
auto = Car(sim=s)
s.activate(auto,auto.run(res=r))
s.simulate(until=100)

Using the existing API, the following program is semantically the same and also works under the OO version:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
from SimPy.Simulation import *
"""Traditional SimPy API"""

## Model components -------------------------------

class Car(Process):
    def run(self,res):
        yield request,self,res
        yield hold,self,10
        yield release,self,res
        print "Time: %s"%now()

## Model and Experiment ---------------------------

initialize()
r = Resource(capacity=5)
auto = Car()
activate(auto,auto.run(res=r))
simulate(until=100)

This full (backwards) compatibility is achieved by the automatic generation of a SimulationXX instance “behind the scenes”.

Models as SimulationXX subclasses

The advanced OO API can be used to generate model classes which are SimulationXX subclasses. This ties a model and a SimulationXX instance together beautifully. See the following example:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
## CarModel.py
from SimPy.Simulation import *
"""Advanced Object Oriented SimPy API"""

## Model components -------------------------------

class Car(Process):
    def park(self):
        yield request,self,sim.self.parking
        yield hold,self,10
        yield release,self,sim.self.parking
        print "%s done at %s"%(self.name, self.sim.now())

## Model ------------------------------------------

class Model(Simulation):
    def __init__(self,name,nrCars,spaces):
        Simulation.__init__(self)
        self.name = name
        self.nrCars = nrCars
        self.spaces = spaces
    def runModel(self):
        ## Initialize Simulation instance
        self.initialize()
        self.parking = Resource(name="Parking lot",unitName="spaces",
                                capacity=self.spaces,sim=self)
        for i in range(self.nrCars):
            auto = Car(name="Car%s"%i, sim=self)
            self.activate(auto, auto.park())
        self.simulate(until=100)

if __name__=="__main__":

    ## Experiment ----------------------------------

    myModel = Model(name="Experiment 1", nrCars=10, spaces=5)
    myModel.runModel()
    print myModel.now()

class Model here is a subclass of Simulation. Every model execution, i.e. call to runModel, reinitializes the simulation (creates an empty event list and sets the time to 0) (see line 24). runModel can thus be called repeatedly for multiple runs of the same experiment setup:

if __name__=="__main__":

    ## Experiments ---------------------------------

    myModel = Model(name="Experiment 1",nrCars=10,spaces=5)
    for repetition in range(100):

    ## One Experiment -------------------------------

        myModel.runModel()
        print myModel.now()

Model extension by subclassing

With the advanced OO API, it is now very easy and clean to extend a model by subclassing. This effectively allows the creation of model libraries.

For example, the model in the previous example can be extended to one in which also vans compete for parking spaces. This is done by importing the CarModel module and subclassing Model as follows:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
## CarModelExtension.py

## Model components -------------------------------

from CarModel import *

class Van(Process):
    def park(self):
        yield request,self,sim.self.parking
        yield hold,self,5
        yield release,self,sim.self.parking
        print "%s done at %s"%(self.name,self.sim.now())

## Model ------------------------------------------

class ModelExtension(Model):
    def __init__(self,name,nrCars,capacity,spaces,nrTrucks):
        Model.__init__(self,name=name,nrCars=nrCars,spaces=spaces)
        self.nrTrucks = nrTrucks

    def runModel(self):
        self.initialize()
        r = Resource(capacity=self.resCapacity,sim=self)
        for i in range(self.nrCars):
            auto = Car(name="Car%s"%i,sim=self)
            self.activate(auto,auto.park())
        for i in range(self.nrTrucks):
            truck = Van(name="Van%s"%i,sim=self)
            self.activate(truck,truck.park())
        self.simulate(until=100)

## Experiment ----------------------------------

myModel1 = ModelExtension(name="Experiment 2",nrCars=10,spaces=5,nrTrucks=3)
myModel1.runModel()

Let’s walk through this:

Line 5:
This import makes available all the objects of SimPy.Simulation and the ones defined by the CarModel module (class Car and class Model).
Lines 7-12:
Addition of a Van class with a park PEM.
Line 16:
Definition of a subclass ModelExtension which extends class Model.
Lines 17-18:
Initialization of the model class (Model) from which ModelExtension is derived. When subclassing a class in Python, this is always necessary: Python does not automatically initialize the super-class.
Lines 21-30:
Defines a runModel method for ModelExtension which also generates and activates Van objects.

API changes

Module SimPy.Simulation

The only change to the API of module SimPy.Simulation is the addition of class Simulation:

Module SimPy.Simulation:
   ################ Unchanged ################
   ## yield-verb constants --------------------
   get
   hold
   passivate
   put
   queueevent
   release
   request
   waitevent
   waituntil
   ## version constant ------------------------
   version
   ## classes ---------------------------------
   FatalSimerror
   Simerror
   ################ Added ################
   Simulation

Thus, after the import:

from SimPy.Simulation import *

class Simulation is available to a program.

Actually,:

from SimPy.Simulation import Simulation

is sufficient and even clearer.

class Simulation

The simulation capabilities of a model are provided by instantiating class Simulation like this:

from SimPy.Simulation import *

aSimulation = Simulation()
## model code follows

Better OO programming style is actually to define a model class which inherits from Simulation:

from SimPy.Simulation import *

class MyModel(Simulation):
    def run(self):
        self.initialize()
        ## model code follows

myMo = MyModel()
myMo.run()

The self.initialize() is not really necessary, as the Simulation instance is initialized at generation time. If method run for a model (here myMo ) is executed more than once, e.g. for running a simulation repatedly, self.initialize() resets the model to an empty event list and simulation time 0.

Methods of class Simulation

class Simulation has these methods:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
class Simulation:
  ## Methods ----------------------------------
  __init__(self)
  initialize(self)
  now(self)
  stopSimulation(self)
  allEventNotices(self)
  allEventTimes(self)
  activate(self, obj, process, at='undefined', delay='undefined', prior=False)
  reactivate(self, obj, at='undefined', delay='undefined', prior=False)
  startCollection(self, when=0.0, monitors=None, tallies=None)
  simulate(self, until=0)

The semantics and parameters (except for self) of the methods are identical to those of the non-OO SimPy.Simulation functions of the same name. For example, to get the current simulation time of a Simulation object so, the call is:

tcurrent = so.now()

Module SimPy.SimulationTrace

The only change to the API of module SimPy.SimulationTrace is the addition of class SimulationTrace:

Module SimPy.SimulationTrace:
  ################ Unchanged ################
  ## yield-verb constants --------------------
  get
  hold
  passivate
  put
  queueevent
  release
  request
  waitevent
  waituntil
  ## version constant ------------------------
  version
  ## classes ---------------------------------
  FatalSimerror
  Simerror
  Trace
  ################ Added ################
  SimulationTrace

class SimulationTrace

The simulation capabilities of a model with tracing are provided by instantiating class SimulationTrace like this:

from SimPy.SimulationTrace import *

aSimulation = SimulationTrace()
## model code follows

Again, better OO programming style is actually to define a model class which inherits from Simulation:

from SimPy.SimulationTrace import *

class MyModel(SimulationTrace):
    def run(self):
        self.initialize()
        # model code follows

myMo = MyModel()
myMo.run()

class SimulationTrace is a subclass of Simulation and thus provides the same methods, albeit with tracing added.

The semantics and parameters of the methods are identical to those of the non-OO SimPy.SimulationTrace functions of the same name.

Methods and attributes of class SimulationTrace

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
class SimulationTrace:
   ## Methods ----------------------------------
   __init__(self)
   initialize(self)
   now(self)
   stopSimulation(self)
   allEventNotices(self)
   allEventTimes(self)
   activate(self, obj, process, at='undefined', delay='undefined', prior=False)
   reactivate(self, obj, at='undefined', delay='undefined', prior=False)
   startCollection(self, when=0.0, monitors=None, tallies=None)
   simulate(self, until=0)
   ## trace attribute ---------------------------
   trace
Attribute trace

An initialization of class SimulationTrace generates an instance of class Trace. This becomes an attribute trace of the SimulationTrace instance.

Trace methods

The semantics and parameters of the Trace methods are identical to those of the non-OO SimPy.SimulationTrace trace instance of the same name.

  • trace.start(self)

Example:

s.trace.start()
  • trace.stop(self)
  • trace.treset(self)
  • trace.tchange(self, **kmvar)
  • trace.ttext(self,par)

Example calls (snippet):

from SimPy.SimulationTrace import *
s = SimulationTrace()
s.initialize()
s.trace.ttext("Here we go")

Again, note that you have to qualify the trace instance (see e.g. the last line of the snippet) with the SimulationTrace instance, here s.

Module SimPy.SimulationRT

class SimulationRT

The simulation capabilities plus real time synchronization are provided by instantiating class SimulationRT.

Methods of class SimulationRT

The SimulationRT subclass adds two methods to those inherited from Simulation.

The semantics and parameters of the methods are identical to those of the non-OO SimPy.SimulationRT functions of the same name.

  • rtnow
  • rtset

Example calls (snippet):

from SimPy.SimulationRT import *
class Car(Process):
   def __init__(self):
      Process.__init__(self, sim=self.sim)
   def run(self):
      print self.sim.rtnow()
      yield hold,self,10

class SimulationStep

The simulation capabilities plus event stepping are provided by instantiating class SimulationStep.

Methods of class SimulationStep

The SimulationStep subclass adds three methods to those inherited from Simulation.

The semantics and parameters of the methods are identical to those of the non-OO SimPy.SimulationStep functions of the same name.

  • startStepping
  • stopStepping
  • simulateStep

Example call (snippet):

from SimPy.SimulationStep import *
s = SimulationStep()
s.initialize()
s.simulateStep(until=100, callback=myCallBack)

Classes with a SimulationXX attribute

All SimPy entity (Process, Resource, Store, Level, SimEvent) and monitoring (Monitor, Tally) classes have time-related functions. In the OO-API of SimPy, they therefore have a .sim attribute which is a reference to the SimulationXX instance to which they belong. This association is made by providing that reference as a parameter to the constructor of the class.

Important

All class instances instances must refer to the same SimulationXX instance, i.e., their .sim attributes must have the same value. That value must be the reference to the SimulationXX instance. Any deviation from this will lead to strange misfunctioning of a SimPy script.

The constructor calls (signatures) for the classes in question thus change as follows:

class Process

Process.__init__(self, name = 'a_process', sim = None)

Example 1 (snippet):

class Car(Process):
    def drive(self):
       yield hold,self,10
       print "Arrived at", self.sim.now()

aSim = Simulation()
aSim.initialize()
c=Car(name="Mine", sim=aSim)

Example 2, with an __init__ method (snippet):

class Car(Process):
def __init__(self,name):
Process.__init__(self,name=name, sim=self.sim)

aSim = Simulation() aSim.initialize() c=Car(name=”Mine”, whichSim=aSim)

class Resource

Resource.__init__(self, capacity = 1, name = 'a_resource', unitName = 'units',
             qType = FIFO, preemptable = 0, monitored = False,
             monitorType = Monitor,sim=None)

Example (snippet):

aSim = Simulation()
aSim.initialize()
res=Resource(name="Server",sim=aSim)

classes Store and Level

Store.__init__(self, name = None, capacity = 'unbounded', unitName = 'units',
            putQType = FIFO, getQType = FIFO,
            monitored = False, monitorType = Monitor, initialBuffered = None,
            sim = None)
::
Level.__init__(self, name = None, capacity = ‘unbounded’, unitName = ‘units’,
putQType = FIFO, getQType = FIFO, monitored = False, monitorType = Monitor, initialBuffered = None, sim = None)

Example (snippet):

aSim = Simulation()
aSim.initialize()
buffer = Store(name="Parts",sim=aSim)

class SimEvent

SimEvent.__init__(self, name = 'a_SimEvent', sim = None)

Example (snippet):

aSim = Simulation()
aSim.initialize()
evt = SimEvent("Boing!", sim=aSim)

classes Monitor and Tally

Monitor.__init__(self, name = 'a_Monitor', ylab = 'y', tlab = 't', sim = None)
::
Tally.__init__(self, name = ‘a_Tally’, ylab = ‘y’, tlab = ‘t’, sim = None)

Example (snippet):

aSim = Simulation()
aSim.initialize()
myMoni = Monitor(name="Counting cars", sim=aSim)