Skip to content

Getting Started

Abel Gómez edited this page Mar 4, 2021 · 40 revisions

This section shows what the Simulation Tool looks like from the users' point of view, and provides a quick description on how to use it in combination with GreatSPN.

UML Modelling

First, it is worth to recall that the modeling phase is done using Papyrus. Since there exists extensive documentation on how to use this tool to create profiled UML models, we will not provide details on the usage of this specific tool. Next figure shows a general view of the Papyrus modeling perspective. On the left of the figure, the different explorers (Project Explorer, Model Explorer and Outline) are shown. The rest of the figure shows the Model Editor and the Properties view. The model itself is depicted in the canvas of the Model Editor.

General view of the Papyrus modeling perspective in Eclipse/DICE-IDE

MARTE Profiling

Profiles, stereotypes and tagged values are defined using the Properties view (Profiletab). The following images show in the Properties view some tagged values that are applied to some model elements. Specifically:

  • The first figure shows the inserDB element, stereotyped as GaStep, and its host demand tagged value. The latter defined as (expr=$timeAdd,unit=ms,statQ=mean,source=est), where $timeAdd is an input parameter representing a mean time duration in milliseconds ms.

Host demand tagged value of the insertDB element, stereotyped as GaStep

  • The second figure shows the the selected control flow, stereotyped as GaStep, and its prob tagged value. The latter is defined as (expr=1-$probActLoop), where $probActLoop is an input parameter.

Prob tagged value of  the selected control flow, stereotyped as GaStep

  • The third figure shows the start element (the initial node), stereotyped as GaWorkloadEvent, and its pattern tagged value. The latter is defined as open=(arrivalRate=(expr=$arrRate,unit=Hz,statQ=mean,source=est)), that is an open workload characterized by a mean arrival rate input parameter ($arrRate), where the rate unit is Hz.

Pattern tagged value of the start node, stereotyped as GaWorkloadEvent

Tagged values are specified in Papyrus-MARTE using the so-called Value Specification Language. As already seen in the previous figures, we can specify model input parameters ($timeAdd, $probActLoop and $arrRate) that will be set to actual values in the simulation configuration step (next section).

Performance analysis

A SSH connection to the simulation server must be provided before launching the simulation (see the Configure GreatSPN SSH Connection section in the First-Steps page). The simulation server hosts the GreatSPN tool, which is in charge of computing the performance metrics selected by the user. The SSH connection is configured in Eclipse - Preferences - Simulation Tools - SSH connection.

Simulation configuration

The first step to be carried out is to set the configuration for the simulation experiments. Therefore, we need to open the Run Configurations... window either by selecting the Run as -> Run Configurations... option from the contextual menu associated to the UML model (Project Explorer view) or by clicking the clock button (red marked in the figure below).

Selecting the Run as -> Run Configurations... option

In this case we can create a new DICE Simulation configuration from scratch from the Run Configuration... windows:

Clicking the the clock button option

In this case a DICE Simulation configuration is created (called Model) that is filled with the information retrieved from the UML model profiled with MARTE:

Note that, depending on the annotations defined in the UML model, the Model configuration can be partially filled or complete. In the running example it is complete and a simulation experiment can be run without any changes.

The figure above shows the Main tab of the Modelconfiguration where it is possible to select:

  • The Model to Analyse, i.e., a UML model annotated with MARTE, by browsing in the worksapce
  • The Active scenario, by selecting the possible GaScenario in the model (the running example includes one one)
  • The NFP to calculate, i.e., the type of analysis: Performance or Reliability

Moreover, the tables shown at the bottom of the Main tab can be used to customize the values assigned to the input parameters specified in the UML model.

The Filters tab of the Model configuration is shown in the following figure:

It includes two panels:

  • The Measure panel, where it is possible to select/deselect the metrics to be estimated during the simulation experiment: such metrics, like the input parameters, are retrieved from the UML model annotated with MARTE.
  • The Sensitivity analysis panel, where it is possible to select/deselect the input parameter configurations. Observe that the tool generates all the possible parameter configurations from the range of values assigned to the input parameters in the Main tab.

The Parameters tab of the Model configuration is shown in the following figure:

It includes General and Simulation parameters of the GreatSPN simulator. In particular, the following are relevant for controlling the duration of a simulation run:

  • Maximum simulation execution time, a simulation run lasts at most the time value set to this parameter
  • Confidence level, the level of the confidence intervals computed for the metrics of the performance model
  • Accuracy, the accuracy of the estimated metrics of the performance model. It is expressed as percentage and it is an integer number (lower the value, higher is the accuracy).

This tab also shows the path of the GreatSPN simulator executable in the simulation server (WNSIM File Path).

Remember to save the changes in a tab of the Model configuration windows by clicking the Apply button before launching a simulation experiment.

Running a simulation experiment

A simulation experiment is run by clicking the Run button in the Modelconfiguration window and it consists of as many simulation runs as the number of configurations selected in the Filters tab. In the running example, a simulation experiments consists of 10 simulation runs.

The simulation can be monitored using the DICE Simulation perspective that can be set by clicking the clock button (red marked in the figure below):

The following figure, shows the DICE Simulation perspective while the simulation experiment is running.

In the figure, three key views can be identified:

  • The Debug view that shows information about the Simulation process (identifier, state, exit value, etc.);
  • The Console view that shows the messages that the simulation process dumps into the standard out and the standard error streams. In the case of GreatSPN, these messages enable to monitor the accuracy achieved by the running process and the number of simulation steps that have been already performed. If an error happens during the process of simulation, it will be notified in the Console view.
  • The Invocation Registry view that shows the starting/ending times and the status of the simulation runs belonging to the simulation experiments.

In the DICE Simulation perspective it is also possible to stop the simulation process at any moment by using the Stop button of the GUI (red marked in the figure). When the simulation finishes, the user can still access to the simulation console and the simulation process information (until he/she cleans the Console view using the corresponding button Detail of the stop button, which force terminates a simulation).

As the next image shows, all the simulation runs terminated correctly (exit value 0) but second-last one that terminated with exit value -10, meaning that the simulation run reached the maximum simulation execution time without achieving the accuracy for all the estimated metrics in the performance model (remember that the maximum simulation execution time and the accuracy are two parameters set in the Parameters tab of the Modelconfiguration window). Therefore, the simulation results are not saved for such run.

Simulation results

The results of a simulation experiment are reported both in textual and graphical formats.

Performance metrics of a simulation run

From the Invocation Registry view it is possible to see the estimated performance metrics by right clicking on a particular simulation run and selecting from the contextual menu the Open Simulation Result option, as shown in the following figure:

A new view, labeled with the id of the simulation run, pops up above the Invocation view as shown in the following figure:

For each performance metric, selected in the Filters tab of the Model configuraton window, the estimated mean value is shown.

Performance curves of the simulation experiment

When a simulation experiments consists of a set of simulation runs, then it is possible to generate 2D plots showing the trends of the estimated performance metrics against an input parameter, in the range of values set during the configuration step.

To generate the 2D plots, we consider again the Invocation Registry view and right click on the simulation experiment as shown in the following figure:

The contextual menu shows the Plot Results... options that launches a wizard for the plot generation. The following three figures shows the windows that pop ups in the wizard:

When the three steps are completed the plot file (data.plot) is saved in the project and a new view pops-up with the 2D plots. The figure below, shows the 2D plot of the utilization metric vs/ the arrival rate. The system is clearly not stable for arrival rates greater than 0.14 Hz, this is a reason for the possible long simulation runs that may occur for such parameter configurations.