GPSS World has a versatile statistics collection capability. For uncomplicated simulations, the statistics in the Standard Report and that reported by the ANOVA library procedure will meet the needs of most users. In these cases, GPSS World will collect and report the simulation statistics automatically. This is discussed in Chapters 11 and 8.
If you need more detailed statistics, the use of variable entities and PLUS Procedures provides excellent control. In these cases, the power of the mathematical library functions is at your disposal. You then define statistics tables in one or more TABLE or QTABLE Statements. The actual recording of data is triggered by one or more TABULATE Blocks in the simulation. Each time a TABULATE is entered, a datum is registered in the table. This can be viewed dynamically in one or more Tables Windows. Alternately, you could record values in Savevalue or Matrix Entities just prior to the completion of the simulation. Tables, Qtables, Savevalues, and Matrices are all reported automatically in the Standard Report, if you have checked them in the Report Page of the Model Settings Notebook.
QTABLE Commands provide an easier way to collect statistics based on Queue Entities. In this case, the tabulations are triggered by entry into DEPART Blocks instead of TABULATE Blocks.
Normally, you should arrange your simulation so that control values are taken from User Variables and the results are in User Variables when the simulation terminates. The reason for this is that this is how the PLUS Experiments created by the Automatic Experiment Generators work. If there is any chance you will want to use them, you will need to make these accommodations.
For more in-depth analysis, you can accumulate simulation results in a Result File by using Open, Close, Read, Write, Seek in the form of GPSS Blocks or procedure calls. This can accumulate simulation results in an historical database that can be used with advanced third party statistical analysis software. For analysis of data in a Result Matrix, the built in Multiway ANOVA Procedure is available.
If you are new to simulation, you will notice that when you repeat a simulation in the same session you may get different results due to random effects. Such effects must be carefully distinguished from the real effects caused by new designs in your simulated systems. The ANOVA library procedure provides you with a simple and yet powerful technique for establishing that your results do or do not emerge above the noise level caused by randomness.
In order to measure the amount of random noise, you must repeat your simulations changing nothing except the random number seeds. These repeat runs are called replications and are important for measuring the statistical noise, i.e. the Standard Error, in your experiment. In the simplest scheme, each different design you choose to simulate would be replicated several times. Then the ANOVA library procedure will use the data to detect if the effects of changing the factors of your experiment were significantly greater than the statistical noise level. The results of this analysis will appear automatically in the Simulation Object's Journal Window.
The ANOVA Procedure provides for push button data analysis. It calculates Confidence Intervals and an Analysis of Variance of simulation results. Normally, during an experiment, you fill a special global Matrix Entity called a Result Matrix with the results of the simulations and then pass it as an argument to the ANOVA Procedure. The details of the use of GPSS World's Multiway ANOVA library procedure are discussed in detail in the next chapter in Section 13.5.3.
If you wish to exclude the effects of starting conditions from your final simulation, you should use the RESET Command to begin the measurement period after transient effects have disappeared. The RESET Command is discussed in Chapter 6. Plot Windows are useful for observing the convergence of a simulation to steady state conditions.
A space-time product is a time duration multiplied by a count. Space-time products are used in the calculation of several kinds of useful statistics. For example, to calculate the wages due to a labor force you could use the number of people who worked each given number of hours. To calculate the total hours worked, you could multiply the hours times the number of people to form a space-time product for each number of hours worked. You could then add up all the space time products in order to calculate the total person-hours. Similarly, many statistics reported in the Standard Reports are accumulations of space-time products.
Facility, Queue, Storage, and Userchain entities have SNAs which require space-time products in their evaluation. For example, the average storage in use is a time weighted average which gives more weight to those numbers which have the longest durations. The accumulated space-time product is divided by the simulated time duration in order to calculate the average storage in use.
Facility, Queue and Userchain entities also have space-time products. They are used to calculate average content via a corresponding SNA. The Facility space-time product is the accumulation of total busy time. The other space-time products are equal to the area under the graph of Queue content or Userchain content.
The RESET command may be used to begin a new measurement period. A RESET command sets the space-time product to zero and the total count equal to the current count. This allows new space-time products to be accumulated. In the SNAs based on space-time product calculations, this method of initialization may introduce a small bias on the low side.
The following table shows the SNAs calculated from space-time products.
Table 12—5. Space-Time System Numeric Attributes
SNA ENTITY CALCULATION
FR UTILIZATION (ppt) Facility 1000 times (Space-time product)
divided by (Total Simulated Time)
FT AVERAGE HOLDING Facility (Space-time-product) divided
TIME PER CAPTURE by (Total capture count)
QA AVERAGE CONTENT Queue (Space-time-product) divided by
QT AVERAGE RESIDENCE Queue (Space-time-product) divided by
TIME (Total count)
QX AVERAGE RESIDENCE Queue (Space-time product) divided by
TIME EXCLUDING ZEROS (Total Count)-(Count finding zero)
SA AVERAGE CONTENT Storage (Space-time-product) divided by
SR FRACTIONAL Storage 1000 times (Space-time product)
UTILIZATION divided by
(Capacity) times (Elapsed time)
ST AVERAGE HOLDING Storage (Space-time-product) divided by
TIME (Sum of storage requirements)
CA AVERAGE CONTENT Userchain (Space-time-product) divided by
CT AVERAGE RESIDENCE Userchain (Space-time-product) divided by
TIME (Total count)
[Table of Contents]