(I) The three-phase approach
Several underlying methodologies to simulation modelling exist. The distinction between these alternative modelling approaches lies in the "next event” selection and timing management. (Hooper, 1986a). The three-phase approach developed back in the late fifties by Tocher (1963) is still the most favoured in the UK (Crookes, 1982). This approach considers two types of occurrences :
- bound activities : which are predictable and hence can be scheduled;
- conditional activities : which are dependent upon certain conditions.
The rules of the three-phase method of Tocher's approach are :
1. A-phase : time scan. Find the next time at which one or more bound activities are scheduled to be executed and advance the clock.
2. B-phase : Execution of bound activities found necessary in the previous phase.
3. C-phase : Testing (and execution of the action part) of each conditional activity.
The executive simulation program cycles round these three-phases until termination time. Termi nation time is given by a preset duration or when a predetermined terminating condition is reached.
Crookes (1982), O ’Keefe and Davies (1983), Balmer & Paul (1986) and Crookes & al (1986) show examples in the current literature of the continuing use of this approach. Their cri terion is the ease in modification of the model. This is due to the relationship between indepen dent conditional rules and the three-phase method. This allows for modular programs to be writ ten making the model easier to analyse, comprehend and extend.
(U). Event-Scheduling
The event-scheduling time control procedure selects, from the event list, the event notice having the earliest occurrence, updates the simulation clock to that time, and invokes the corresponding event routine. Any condition testing, other than on clock time must occur within event routines. Events are chosen and processed successively until termination time. (Hooper. 1986a; Laski, 1965; Gordon,1978; Fishman. 1973).
(111). Activity-Scanning
The Activity-Scanning approach chooses the next event based on scheduled time and condi tion testing. The basic concept is an "activity" which is (conceptually) a system state transition requiring a period of time. An activity is usually represented as two distinct events which mark the beginning and the end of the activity. The activity-scanning time control procedure scans activities in priority order for time eligibility and other activation conditions, executing the activity routine of the first component whose activation conditions are met. When an activation occurs, the scan starts over again in priority order. This process continues until termination time. (Hooper. 1986b; Laski. 1965; Gordon.1978; Fishman. 1973).
(iv). Process-Interaction
This approach has characteristics related to both the event-scheduling and activity-scanning approaches. Components of a system progress through a sequence of steps (referred to as a pro cess). Each step may consist of a condition segment and an action segment. Execution of the condition segment determines whether execution of the action segment may occur.
The process-interaction time control procedure has two event lists : a future event list (FEL) containing event notices for events scheduled for execution at a later check time, and a current event list (CEL) containing event notices for events which are already eligible from the stand point of time to be executed, but whose other conditions may not yet have been met. Each event notice contains an indication of its components current step location in a process. When time is advanced, all events scheduled for current time are moved from FEL to CEL. Then a
CEL scan occurs. This consists of evaluating each entry's condition routine to determine whether the corresponding components may move to the next step. If so, the step's action segment is exe cuted. A component moves through as many successive steps as possible (ie as long as time need not advance, and condition segments are found to be true). When a component "stops" (due to time or other conditions), the scan resumes with the next CEL entry. When no CEL component can move time is advanced. (Hooper. 1986a; Gordon. 1978; Fishman, 1973; Zeigler, 1976).
(v). Cellular Simulation
The methodology o f cellular simulation is described in Spinelli & Crookes (1976). The basic idea of cellular simulation is to split the simulation into non-overiapping activity groups (or ceils). Each cell can then be considered as a simulation in its own right.
Computational efficiency is the main advantage of this approach. Indeed, in terms of the three-phase model, a great saving in the execution time is possible, because there will be no test ing for a C-activity within a cell unless that cell has had a B-activity executed since the last time the clock was stopped or an elememt has entered or left the cell. This cuts down substantially on the checking of those activity tests which must fail.
Spinelli and Crookes (1976) also indicate the possibility of modelling large systems in separate cells, written at different times and for different purposes and later combined.
2.2.5 Enhancements to the practice of Simulation
Computer simulation is not trivial. There are many difficulties (see Conway & al, 1959; Conway. 1963). Realistic simulation may require long computer programs of some complexity. Simulation verification, validation and experimentation are not easy either. (Pidd. 1984). During the history of simulation, there have been a series of developments to enhance the practice of simulation and to facilitate its usage.
(i). Languages
By 1965. Tocher was able to review a large number of different simulation languages. Most of these were dedicated to a specific machine only and Tocher concluded that the choice of a simulation language would most likely be made by the type of machine available to the user.
The main conclusion from his detailed comparison of these languages was that for occa sional use, a simple language, which is easy to understand and learn may be more valuable than one of the sophisticated languages that has many facilities but are much more complicated to use and understand.
Since this review, the number of simulation "purporting to aid the unwary user " has greatly increased (Crookes. 1982). Crookes counted 137 such languages. Although concerned by this large number, he noted a "level o f model verification not previously attainable by the use o f practical outputs from a running simulation" ( a major benefit of Visual Interactive Simulation (see later)).
In line with the increased use of microcomputers, another contribution to the popularisation of simulation has been the great number of simulation languages and packages implemented on microcomputers. (O’Keefe. 1984).
(11). Packages
(a). Program Generators
Program generators are probably the most obvious example of packages designed to aid the simulation analysts. The best known are CAPS (Clementson. 1974) and DRAFT (Mathewson. 1974). O’Keefe (1984) put forward the following advantages of program generators :
- they free the model builder from a considerable amount of programming;
- the generator enforces a particular modelling approach and can catch errors and incon sistences in using that approach at an early stage in the development cycle.
But critics of generators such as Tocher (see O'Keefe. 1984) argued that generators simply han dle the parts of the model that are easy to code such as queue manipulation, leaving the more difficult part to the user. Flitman (1986) concluded that "although a bold move toward the automated programming of simulations, providing a package which does the work for you (though not necessarily in the same way you would) can only be the start o f a research drive in that direction".
Recently, these tools have reappeared in the literature under the description of Interactive Simulation Program Generators (ISPG). See Balmer & Paul (1986).
(b) Interactive Menu Driven Interpreter
A more modem approach to program generators was put forward by O ’Keefe (1984). This Interactive Menu Driven Interpreter allows rapid development and immediate execution of visual interactive simulation models. The model can be tested while it is being written and hence the distinction between building and interaction disappears. The interpreter is programmed using a package o f PASCAL simulation routines.
(iii) Visual Simulation
O'K eefe (1984) noted four ways by which a simulation model could be enhanced and which P.C. Bell (1985) divided into two categories :
Representational Graphics
1. providing a trace : values of selected variables are displayed over time.
2. displaying time series or histograms
3. displaying tables of data which are updated by the simulation (Brown. 1978).
Ionic Graphics
4. displaying a picture corresponding to the real life system being modelled in ordered to provide an animation of this real world, (see Palme (1977).
The advantages of visual simulation (Crookes, 1982) are :
- it aids in program verification by allowing the writer to determine whether the program he has written is the program he thought he had written.
- it aids in program validation because it helps the user to believe the computer program to be a fair representation o f his real world.
(iv) Visual Interactive Simulation (VIS)
Hurrion conceived the term VIS. His PhD. in 1976, led to the development of VISION (Hurrion 1980), which in turn led to the Fortran based SEE-WHY system. The basic design cri teria of VIS are:
1. a simulation language in which it is possible to write complex industrial problems. VIS does not constrain the user to a particular modelling approach : he may choose between event, activity or cellular-based structure (Hurrion, 1980).
2. the ability of a 1-1 correspondence between elements in the model and elements as described visually on visual display units (VDU).
3. flexibility at run time. This is th e possibility to interact:
- basic interactions which allow inspection,
- structural interactions which allow changes.
This, in effect, permits the user to see what is happening. "The visual interactive approach has ceased in making simulation a 'b la ck box' technique, but now opens up the method for manage ment to look inside. It now becomes a transparent box which greatly assists in the problem of communication and model credibility. (Hurrion, 1980)" This has made validation so much easier, that Rietz (1983) claims th at it has become indispensable.
Indeed by watching a simulation model progress through time, and having the ability to interact with it. a user can improve his analysis and understanding of the original problem situa tion (Hurrion and Seeker, 1978). This fact was highlighted by P.C. Bell (1985).
The technique of VIS was extended by Hurrion and his research students at Warwick University (Seeker, 1977; Brown, 1978; Rubens, 1979; Withers. 1981; Moreira da Silva, 1982; Flitman. 1986) during a series of both internal and joint projects. The main conclusions of this work was ;
1. The visual aspect has a wide appeal (Brown, 1978).
2. Interaction with the model increases confidence in it. Users feel that they are "partici pants rather than spectators" (Brown, 1978).
3. "Situations may arise that the decision-maker may never have envisaged" (Brown, 1978). The picture captures this. Such extreme situations can be lost in the aggregate out put from a batch simulation.
4. "There is a need for development o f good interfaces that enable the decision maker to use his creative thinking and pattern-recognition capacities to their maximum potential"
(Moreira da Silva, 1982).
Other users of VIS have reported similar findings (see Crookes 1982, Ranyard and Blewitt, 1983).
A traditional problem with simulation is that it has not been possible to change underlying logic of the simulation whilst it is running. For exam ple, in coding logic concerning movement o f entities about a simulation, the analyst is setting this logic within the procedural high-level language. In an academic environment, Flitman (1986) addressed this problem by separating the logical aspect form the execution aspect. A second approach to this problem has been presented by Istel (1987) who has produced a package called W ITNESS that is data driven.