RADAR - The Requirements and Architecture Decision AnalyseR

RADAR is a lightweight modelling language and automated decision analysis tool to support multi-objective decision under uncertainty in requirements engineering and software architecture.

RADAR analysis provides useful feedback to requirement engineers and software designers about:

Which decisions are better than others in the decision model of a particular software intensive system.
What objective values can be attained with different designs.
What tradeoffs can be made between shortlisted designs (solutions).
What parameter uncertainty may deserve additional data collection and analysis before making their decision.
What parameter uncertainty does not matter to their decision.

Papers

The RADAR modelling language and decision analysis support tool are described in the following research papers for publication:

RADAR: A Lightweight Tool for Requirements and Architecture Decision Analysis, published and presented published and presented in International Conference on Software Engineering (ICSE) 2017, Buenos Aires, Argentina.
Authors: Saheed A. Busari and Emmanuel Letier

Towards search-based modelling and analysis of requirements and architecture decisions, published and presented in International Conference on Automated Software Engineering (ASE) 2017, Urbana Chmapaign, Illinois, United States of America.
Authors: Saheed A. Busari

Modelling and Analysing Software Requirements and Architecture Decisions under Uncertainty, PhD Thesis
Authors: Saheed A. Busari

The scalability analysis of RADAR desision support tool is described in the paper below:

Scalability Analysis of the RADAR Decision Support Tool.
Authors: Saheed A. Busari and Emmanuel Letier

Download

The RADAR tool is available and can be downloaded here, and all model examples used in the paper can be downloaded here.

We have also made the source code of the tool available in GitHub. The tool is implemented in Java and uses ANTLR to generate model parser, generates diagrams in DOT format , and uses Graphviz (also available in browser ) to visualise the diagrams.

How to Use RADAR

RADAR is a self-contained jar file. Simply download the file to your computer and double click to launch the application (needs JRE 1.7 or later versions). Once the application is launched successfully, a window appears with three panels as shown in figure 1:

RADAR has three main tabs that displays by default when it is launched:

Editor where modellers can write their own models and also load existing models for review.
Analysis Result where the results of the optimisation and information value analysis are displayed.
Console where the RADAR analysis status are written.

The Analysis Settings , shown in figure 5, is used to specify parameters for model analysis and the path where Graphviz is installed to enable visualisation of the variable AND/OR graph and decision dependency graph. Examples of these parameters includes:

The number of Monte-Carlo simulation run.
Information value objective to be used for computing the expected value of total and partial perfect information (evtpi and evppi)
Variable sub-graph for which the AND/OR graph is generated from the specified variable.

Other tabs that are displayed after a successful RADAR analysis are AND/OR goal graph (figure 3), decision dependency graph (figure 4) and the Pareto-front (figure 6)

To analyse an existing model, the following steps can be followed:

Open the RADAR file (we recommend starting with the first shown in figure 1 i.e. refactoring cost-benefit analysis) by simply clicking on the file menu and then click Examples to load the existing model on the editor. if successful, you will see the model displayed in the model board. At this point, users can edit the model and save changes by clicking on save under the file menu.
Go to the Action menu and click parse model to check that the specified model conforms to RADAR syntaxes defined in the paper. If not, an error message is displayed. If successful, you will be prompted to either continue with analysing the model or you could decide to further review the model and later analyse the model by clicking analyse model under the Action menu. Note that before parsing the model, you will be required to specify the output directory, which stores model analysis results.
If you click continue with model analysis, RADAR analyses the model as described in the paper, and the analysis results, such as the optimisation analysis, Pareto front and information value analysis (if the information value objective is specified), are displayed in another tab as shown in figure 2.

RADAR also saves the following in the directory where the jar is located:

The model analysis results in .csv and .out extensions.
The AND/OR variable dependency graph (which helps to communicate and validate traceability links between technical software characteristics) and decisions dependency graph (which helps to visualise the dependencies between decisions and options, and also to view a potentially large design space in terms of a smaller set of decisions and options). These graphs are saved in DOT format and can be viewed using Graphviz (also available in browser). For RADAR to generate these graphs, users have to download and install Graphviz and then specify the path of the executable under 'Graphviz Settings' which can found in the 'Analysis Settings' tab.
The Pareto front plot (in .PNG format) which is the set of optimal solutions that show the trade-off that can be made between the multiple and conflicting objectives.

Figure 1: RADAR refactoring model example

Figure 2: RADAR refactoring model analysis result

Figure 3: RADAR refactoring model AND/OR graph.

Figure 4: RADAR refactoring model decision dependency graph

Figure 5: RADAR Analysis Settings page.

Figure 6: RADAR refactoring model Pareto front.

To analyse your own model, simply follow the same steps after having edited your model in the tool or using an external text editor.

For any question, email: {saheed.busari.13, e.letier}@ucl.ac.uk

Examples

We have applied the tool to the following examples.

Example	Brief Description	Report	Radar File
Refactoring cost-benefit analysis	The small example used in the paper to introduce the language and decision analysis method	refactoring	refactoring.rdr
Plastic card fraud detection system	The running example used in the paper to illustrate the language and analysis functions. The problem consists in selecting among alternative design for a fraud detection system so as to minimise financial loss due to fraud and minimise the number of fraud alerts to be investigated manually	fraud-detection	fraud-detection.rdr
Document sharing security policy	This example analyses security policy decisions for an organisation concerned with leaks of confidential information. The analysis extends previous models developed by Tristan Caulfield and David Pym for a real organistion.	sharing-policy	sharing-policy.rdr
Emergency Response System	This example illustrates the use of RADAR on an architecture decision problem used in previous publications (see the report for details). The design space for this model includes 6912 solutions and takes around 2 minutes to analyse.	emergency-response	emergency-response.rdr
Satellite Image Processing	This example was first introduce to illustrate the CBAM architecture decision method (see report for full references). We show here how to analyse this problem in RADAR and compare RADAR's approach to CBAM.	nasa-ecs	nasa-ecs.rdr
London Ambulance Example	This example illustrates the design decisions of the London Ambulance Service. We show here how to analyse this problem with RADAR. There are 5 design decisions and the solution space id 648 and the anlaysis was completed in milli seconds.	london ambulance service	las.rdr

We have also applied RADAR on different synthetic models to understand the scalability of the tool. The synthetic models are generated using a model generator that generates random syntactically valid RADAR models with a given number of objectives, decisions, number of options per decisions and minimum number of model variables. To create a synthetic model of a specific size, download the model generator jar file and run the command below:

java -jar ./Downloads/SytheticModelGenerator.jar ./Downloads/SyntheticModel.properties

In the command above, SytheticModelGenerator.jar is the program that generates the synthetic model and SyntheticModel.properties stores values for the RADAR model constructs, such as the number of objectives, decisions, number of options per decisions and minimum number of model variables. The file SytheticModel.rdr contains a synthetic RADAR model with decision dependency and has 2 objectives, minimum of 10 model variables, 4 decisions with their corresponding options ranging between 2 and 3.

The RADAR files can be opened and analysed in the tool following the steps outlined above.

Roadmap

Radar is an early prototype that we are constantly improving. Our current development goals include:

Applying RADAR to additional examples.
Extending the language to support decisions with non-mutually exclusive options.
Provide fully documented API for use of Radar by external tools.
Facilitating generation of decision analysis report by embedding RADAR model fragments and analysis results within markdown documents.