Discrete Event Simulator

From Mobius Wiki
Jump to: navigation, search


The simulator can be used to solve models using discrete event simulation. It takes the model and parameters specified by a study, links the libraries for those models together with the simulator library, and runs the executable. The simulator can be used to solve any model specified in Möbius, and can be used to solve for transient and steady state results.

Simulation Parameters

The Simulation Parameters tab on the Simulator editor allows you to specify all of the simulation parameters and is shown in <xr id="fig:sim_parameters" />.

<figure id="fig:sim_parameters">


<xr id="fig:sim_parameters" nolink />: Simulation Parameters tab of the Simulator.

The parameters on the page can be roughly broken down into four categories that proceed down the tab.

  1. Study and experiment selection
  2. Simulation execution parameters
  3. Compilation options
  4. Checkboxes

Study and experiment selection

The top of the tab starts with the Current Study field, which contains the name of the study to be solved. The study is initially set when you create the simulation, and you are prompted to specify the solver child. You can change the study by typing a new study name into the field, or by clicking the Browse button. If you click on the Browse button, the Select Solver Child window will pop up. It shows all of the studies and prompts you to select one.

The Experiment Activator button opens a pop-up window (shown in <xr id="fig:sim_experiment" />) to allow you to view and select the experiments in the study. Recall that all global variable values are assigned in the study, and that each unique set of assignments is called an experiment. The pop-up window shows all of the experiments in the study, the global variable values for each experiment, and the experiments that have been activated in this simulation so that they will be solved. You can activate or deactivate an experiment by checking the box above the experiment. You can activate or deactivate all experiments by clicking on the Activate All or Deactivate All buttons.

<figure id="fig:sim_experiment">

Sim experiment.png

<xr id="fig:sim_experiment" nolink />: Window for enabling and displaying experiment values.

Simulation execution parameters

The next set of parameters affect the execution of the simulation.

  • The radio boxes under Simulation Type specify whether the model solution should be a steady state solution or a terminating solution. If the model contains transient rewards, the Terminating Simulation option will be enabled, and if the model contains steady state rewards, the Steady State Simulation option will be enabled. If all the rewards are either transient or steady state, only one option will be enabled, and it will be selected by default.
  • The pull-down menu Random Number Generator is used to select the pseudo-random number generator to use in the solution. The default option is the Lagged Fibonacci random number generator, but the Tauseworthe random number generator can be used.
  • The Random Number Seed field specifies the seed for the pseudo-random number generator. The default value is 31,415, but any value can be used. Once the seed has been specified, the sequence of numbers is deterministic. Multiple simulations with the same random number seed should generate the same results5.
  • The Maximum Batches field specifies the maximum number of batches (in steady state simulation) or replicas (in transient simulation) that the simulator will run. The simulator will run several batches or replicas to generate data for the confidence intervals. The simulator will continue to run more batches or replicas until all of the results have converged to their specified confidence intervals. This option will terminate the simulation if the simulator has run the maximum number of batches, even if not all of the reward variables have converged to their specified confidence intervals.
  • The Minimum Batches field specifies the minimum number of batches (in steady state simulation) or replicas (in transient simulation) that the simulator will run. The simulator will run several batches or replicas to generate data for the confidence intervals. The simulator will continue to run more batches or replicas until all of the results have converged to their specified confidence intervals. This option ensures that the simulator will run at least the minimum number of batches or replicas before finishing. This is useful in simulating rare events, for which the model’s reward variables may appear to converge too early if no rare events have occurred.
  • The Number of Batches per Data update field specifies how many batches or replicas the simulation executables should complete before sending those results back to the editor. If the simulation is run on multiple machines, each machine will send an update after simulating the specified number of batches or replicas.
  • The Number of Batches per Display update field specifies how many batches or replicas should be received by the editor before it updates the user interface. A low number will provide the most up-to-date data, but can cause the user interface to use too many resources. A larger number will reduce the load created by the user interface, but will also provide fewer updates. In general, this number should always be greater than or equal to the number specified in Number of Batches per update.

5 The simulator is multi-threaded and can be distributed. Due to timing variations in a distributed environment, the results will not necessarily be identical when an experiment is simulated multiple times with the same random number seed, even though the random number sequences will be identical.

Compilation options

The next set of options control how the simulation is compiled and run, allowing for optimizations, trace output, and specification of the run name.

  • The Build Type pull-down menu allows you to set the model to run in either optimized or normal mode. In normal mode, the model is compiled without compiler optimizations, and it is possible to specify various levels of output by changing the trace level. In optimized mode, the model is compiled with compiler optimizations, and all trace output is disabled for maximum performance. You would normally use normal mode for testing a model and making sure it is behaving properly, while you would use optimized mode for production simulations.
  • The Build Architecture pull-down menu allows you to set the model to execute in either 32-bit or 64-bit mode. On 32-bit machines (or machines running a 32-bit operating system), this drop down menu will be disabled and only 32-bit execution is available.
  • The Trace Level pull-down menu allows you to set the level of detail included in a trace of the simulation. The trace includes key information about the evolution of the simulation in a text file, which can be used to debug a model. The available trace levels are:
0: No Output: No trace file is generated.
1: Firing Time and Activity Name: This option will include the simulation time of every activity completion, and the name of the activity that fired.
2: Level 1 plus Minimal State: This level adds some relevant state information from the time of the activity firing to the previous level’s activity firing times and names. Specifically, it shows the values of state variables that are either part of the enabling conditions for the activity, or part of the state affected by the activity.
3: Level 1 plus All State: This level prints out all of the state information at every activity firing, in addition to the name of the activity and the time it fires.
4: All: This level prints out all of the information from the previous levels plus more data on which activities are enabled, become enabled, or become disabled when an activity fires.
The trace-level options can generate useful information that you can use to understand the behavior of your models. However, it can also generate more information that you want. Generally, trace output is only used with a small number of batches.
  • The Run name field specifies the name for this run of the simulation. This is used to name trace files and output files. Change the run name any time you want to save existing results from a previous run of the simulator. Otherwise, the new results will overwrite any existing results.


The last set of options are set using checkboxes. Most of them control the output of the simulator, while one of them controls the variance calculation.

  • Store simulator console output to file stores all of the output from the editor display to a file.
  • The Store observations to ASCII .csv file option can be used to save the raw simulation results to a file for later use. Again, this option is useful if you are running remote simulations over a low-bandwidth connection, because you can access the observations at a later time to get the results of the simulation. The option stores the observations in an ASCII file format. ASCII is human-readable, and can also be loaded into tools like Excel for custom analysis.
  • The Store observations to binary .dat file option can be used to save the raw simulation results to a file for later use. This option is useful if you are running remote simulations over a low-bandwidth connection. You can access the observations at a later time to get the results of the simulation. The option stores the observations in a binary file, which should be smaller than the ASCII file the other option would have produced. The binary files are used by Möbius for off-line processing (see Section 3.3.2).
There used to be a checkbox in the Simulation Parameters tab called Use Jackknife Variance Calculation that made the simulator use the advanced jackknife method of variance calculation. In Möbius 1.8.0 and later, the method is always used; therefore, the checkbox has been removed.

Network Setup

Use the Network Setup tab, shown in <xr id="fig:sim_network" />, to select the computers on which the simulation will be run. You can parallelize simulations by executing multiple replicas of the simulation on different machines at the same time, with near linear speed-up in solution time. On the left of the tab is the Available Systems list of machines or groups of machines that can be used in the solution of the model. You can select one or more machines for use by selecting their names in the list, and then clicking on the >> button. Clicking on the button will transfer all of the selected machines to the Selected Systems list. You can deselect machines in the same manner, by selecting their names in the Selected Systems list and clicking on the << button.

<figure id="fig:sim_network">

Sim network.png

<xr id="fig:sim_network" nolink />: Network Setup tab for the Simulator.

You can edit the list of available systems by clicking on the Edit Machine/Group Info button. Clicking on the button opens the Network Machine and Group Settings dialog, which is described in Section ??.

The Number of Processors per Experiment drop-down menu allows you to select the number of processors to use per experiment. The simulator can assign multiple machines to simulate the same experiment, and each experiment will be assigned a number of machines no greater than the specified number. Machines will be assigned to the lowest-numbered experiment that has not yet been solved and that has fewer than the specified number of machines.

The Maximize Processor Usage checkbox overrides the Number of Processors per Experiment setting. If the Maximize Processor Usage option is enabled, the simulator will distribute experiments to all the machines, such that each experiment is running on a similar number of machines. When an experiment finishes, the machines that were assigned to it are then reassigned to the other experiments. In that way, all of the experiments are solved as quickly as possible.

Running Simulation

The Run Simulation tab, shown in <xr id="fig:sim_run" />, is used to start and stop the simulator. As the simulation is being compiled and run, messages are produced and are shown in the Simulation Status area. The information includes the name of all the files being compiled and linked, and indicates when the execution of the simulation begins.

<figure id="fig:sim_run">

Sim run.png

<xr id="fig:sim_run" nolink />: Run Simulation tab of the Simulator.

On the bottom of the tab are a couple of check boxes and a series of buttons to control the running of the simulation. The checkboxes Skip compile and Force compile control whether the underlying files are compiled before starting the simulations. By default, the simulator editor will recompile the minimum set of files before starting the simulations for the first time if neither checkbox is checked.

There are two modes for generating simulator results. The first mode is on-line, in which the simulation executables are run and the results are computed and displayed by the editor as the solution converges. The second mode is off-line. There are two steps in off-line simulation: the first step is the launch of the simulation executables with the binary output file option enabled, and the second step is the processing of the binary output files when the simulator processes have completed. The two modes are discussed in the following sections.

On-line simulation

The Start Simulation button starts the simulator. Any files that should be recompiled are recompiled, and experiments are assigned to the selected machines.

The Stop Simulation button will stop a running simulation. It will stop all of the simulations started on remote machines, and finish any remaining result computation.

The Client Information dialog, shown in <xr id="fig:sim_clientinfo" />, is used to check the status of a client machine during solution. The Client Information window will pop up with a list of client machines used in this simulation. When you double-click on any of the system names, the Client Status window, shown in <xr id="fig:sim_clientstatus" />, will pop up to provide the name of the system, the architectures, and the operating system.

<figure id="fig:sim_clientinfo">

Sim clientinfo.png

<xr id="fig:sim_clientinfo" nolink />: Pop-up window to check the status of the client machines.

<figure id="fig:sim_clientstatus">

Sim clientstatus.png

<xr id="fig:sim_clientstatus" nolink />: Pop-up window that displays the status of a client machine.

Off-line simulation

The Process Binary File button allows you to process a binary output file that was generated if you checked the Store observations to binary .dat file option, as described in Section 3.1.4. When you press the button, a file browser will open to allow you to select the appropriate binary data output file(s). Once you have selected the files, Möbius reads them and processes the data as if the simulation was running on-line, and displays the results after all data has been read.

Binary data files can be generated through multiple simulation runs. You can launch runs with the Möbius simulation editor or launch the simulation executable directly via the command line.

One reason to use the off-line processing capability is to generate solutions without having to run the Möbius graphical interface. You can launch jobs on the command line (directly or via user-constructed command line scripts), collect the results, and transfer them to a machine with graphical access on which the data files can be processed to compute the results.

Another reason to use off-line processing is to generate more simulation replications than could be generated in one session. Repeated executions of the simulator (using a unique random seed and run name each time) will produce multiple sets of data files. These files can be processed together at one time to generate results based on all the runs.

Simulation Info

The Simulation Info tab displays the results for the experiments as they become available, and is shown in <xr id="fig:sim_results" />. At the top left of the tab is the list of all the enabled experiments and their statuses. The Status column indicates whether the experiment has not started, is running, or has completed. The # CPUs column indicates how many processors are currently assigned to the given experiment, and the Batches column indicates how many batches or replicas of that experiment have completed and been reported back to the user interface. You can select an experiment by clicking on its row in the table. Selecting the experiment in the list updates the Selected Experiment display in the lower half of the window. The Selected Experiment display shows the simulation progress by reporting the number of replications computed and showing the mean and variance values and confidence intervals for each reward variable. When the result converges so that the computed confidence interval is within the tolerances specified in the Simulation panel of the Performance Variable Editor (see Section 6.1.6 of Building Models), the color of the reward variable row changes from red to blue.

<figure id="fig:sim_results">

Sim results.png

<xr id="fig:sim_results" nolink />: Simulation Info tab of the Simulator.

There are three buttons to the right of the experiment list:

  • The Terminate Selected Experiment button stops the solution of the currently selected experiment. All machines that are simulating that experiment stop doing so, and are reassigned.
  • The Terminate All Experiments button stops the solution of all the experiments. The simulator receives the final updates and displays the final results.
  • The Show Results button opens the Simulation Results window, which is discussed in the next section.

Simulation Results

The results of the simulation are displayed in the Simulation Results tab when the simulation finishes or the user clicks on the Show Results button of the Simulation Info tab. The window displays the human-readable text output created by the simulator. The file includes all of the parameters for the model, all of the experiments, and all of the results. In addition to the human-readable output, a comma-separated-value file (.csv) is also created. The csv file is designed to be simple to import into third-party plotting and analysis tools (such as Excel).

Command Line Options

The simulator is normally run from the simulator editor. However, it can also be run from the command line. The simulator executable resides in the project tree, in a path following this pattern: <Möbius Project Root>/<Project Name>/Solver/<Solver Name>/. It can be run with the command line options listed in <xr id="tab:command_options" />. These options will be shown if you execute the simulator on the command line, without any options.

<figtable id="tab:command_options">

<xr id="tab:command_options" nolink />: Command line options for the Möbius simulator.
Option Description
-a <name> Write ASCII output to file <name>.
-b <name> Write binary output to file <name>.
-c <num> Client identification number (default = 0).
-e <num> Experiment to run, “Experiment 1” is num=0 (default = 0).
-f <name> Trace file name (default=<stdout>).
-l <num> Trace level (0-off, 1,2,3,4 [default = 0]).
-n <num> Num observations sent in 1 update to simulator editor (default = 1000).
-N <num> Maximum number of batches to run (default = infinite).
-p <num> Communication port of network server (default = 10000).
-r <num> Random number generator

(0 = lagged Fibonacci [default], 1 = Tauseworthe).

-s <num> Random number seed (31,415 [default]).
-t <num> Simulation type (1 = terminating [default], 0 = steady state).
-w Simulator will wait to send output until the socket connection has been established. Used when the simulator is launched by the Möbius editor.