Tutorial: An Object Oriented Approach to Using SimPy

Authors:K G Muller Date:2010 April

Introduction

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.

SimPy, on the other hand, supports 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.

Object-oriented model implementation allows 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.

The most effective use of the object-oriented approach is an iteration over Object Oriented Analysis, Object Oriented Design, and Object Oriented Programming.

Simulation studies are typically performed to study systems to understand the relationships between its components or to predict how the system will perform in a changed environment. They are accomplished by building a model of a system and experimenting with it.

In modelling, it is only necessary to consider aspects of the system that affect the problems being investigated.

Identifying those aspects is rarely trivial and often requires trials with a model and subsequent model refinement.

This short tutorial will attempt to show how this can be done for simulation modelling with SimPy. It is no comprehensive course on the object oriented approach, though. There are many publications on the web and also books teaching OO in general.

Object Oriented Analysis

The goal of OO analysis is to identify the scenary in which the system to be modelled operates and the system components in terms of objects, their attributes, actions and interactions.

A good start is to write a concise scenario description in natural language and to look for terms which identify objects, attributes, etc.

Unfortunately, there is no algorithm for this, only heuristics. For english language analysis, the article Natural Language Analysis for Domain and Usage Models gives a good introduction.

Here is a useful set of heuristics for mapping parts of speech to model components:

Part of speech	Model component	Examples
Proper noun	Object	Cashier
Common noun	Class	Bank customer; clerk
Doing verb	Operation	visit; withdraw money
Being verb	Inheritance	Is a clerk
Having verb	Aggregation	Has an account
Modal verb	Constraints	Clerk must be at counter
Adjective	Attribute	Number of clerks

It is essential that the scenario description be done in the terms of the user and/or the system domain.

Here are further useful heuristics for identifying objects:

• recurring nouns (e.g. transaction)
• real-world entities the model should include (e.g. a customer, cashier, bank account)

Model developers name and briefly describe the objects, their attributes, and their responsibilities as they are identified. Describing objetcs, even briefly, allows simulation modellers to clarify the concepts they use and avoid misunderstandings. Initially, modellers need not, however, spend a lot of time detailing objects or attributes. They will be refined during the unavoidable iterations and revisions. At the end of the analysis process, this should result in a stable and sufficiently detailed statement on objects and attributes. Such a statement is esential for gathering simulation inputs from users and mapping simulation results into real world terms.

Here is a very simple scenario description:

A bank has a number of counters staffed by clerks. It also has a number of Automatic Teller Machines (ATMs). During the bank’s opening hours, customers visit the bank at different times to perform one or more transactions requiring service by a bank clerk at a counter or use of an ATM. All service by clerks is provided at counters. Counters can be closed and unstaffed. If the clerk or ATM they need is busy, they wait for him/her/it to become available. After having performed all their transactions, they leave the bank. The waiting times of the customers and the load on clerks and the ATMs should be estimated by using a simulation model.

The highlighted words and the scenario text suggest objects, attributes, operations and constraints:

Objects: bank, counters, clerks, customers, ATM, transactions, service

Attributes:

Of bank:

counters, clerks, ATMs, opening hours

Of customers:

arrival times, transactions, waiting times, departure times

Of counters:

status(staffed, closed)

Of clerks: availability, load

Of ATMs:

availability, load

Operations:

By customers:

arrive, perform transaction, wait, leave

By clerks:

staff counter, provide service

By ATM:

provide service

Constraints:

Counter service:

if clerk available

ATM service:

if ATM available

Customer departure:

after having performed all transactions

Customer arrival:

if bank open

It is highly unlikely that these initially identified objects, attributes etc. are either sufficient or all necessary for the intended simulation model. They do provide a starting point, though, for seeking further details (e.g. by interviews of staff with domain knowledge) and building a first rough object design.

Object Oriented Design

The next step is to develop an initial high-level object design from the results from the OO analysis. This means that for each object identified, a class must be defined to which the object belongs, i.e. a generalization of the object. The class encloses all the properties of an object, i.e. attributes and operations. The attributes define all the data members of an object. The behaviors define how the object interacts with other objects and changes its own attributes.

This class modelling should be done even if there is only one object of this class in the scenario being modelled. It should be noted that the term class here is not to be confused with the class construct in Python (and therefore SimPy). The class here is just the description of one or more similar objects. It will become obvious in the following sections that the availability of class in Python, the implementation language being used for SimPy models, is a great benefit. It allows clear, relatively simple mapping from the OO design to a SimPy program.

Class Bank:

Attributes:

counters, clerks, ATMs, opening hours

Operations:

open, close

Class Customer:

Attributes:

arrival time, transactions, waiting time, departure time

Operations:

arrive, get service for a transaction, leave

Class Counter:

Attributes:

staffing(staffed, closed)

Operations:

open

Class Clerk:

Attributes:

availability (available, busy, absent)

Operations:

provide service, staff counter

Class ATM

Attributes:

availability(available, busy)

Operations:

provide service

Control object needed to set up simulation experiment:

Class Model:

Attributes:

model components, experiment data

Operations:

get experiment data, generate component objects,

run model, report results

Object Oriented Programming

Object Oriented Building of SimPy Simulation Models

Basic Recipe

• Describe the situation to be modelled in natural language.
• Identify the nouns and verbs as candidates for classes, objects and actions, respectively.

Model Components Provided By SimPy

User-Developed Components

SimPy Model Structure

Re-use of Components and Models

Examples: The Bank World

Objects and Activities

Bank Model Components

Model 1: ....

Model 2: ......