PistarGODA MDP

This artifact extends GODA framework into supporting the goal modeling of SAS under multiple classes of uncertainty (named uncertainty related to the system itself, to system goals, and the environment). From the goal model, the framework automatically generates: (1) reliability and cost parametric formulas parameterized with uncertainty; and (2) a Markov Decision Process (MDP) of the system alongside PCTL properties. The formulas express the modeled system's overall reliability and cost. They are used for runtime verification and to guide the sysnthesis of adaptation policies in SAS (see article). The MDP and PCTL files are used for probabilistic model checking (in which we refer to the PRISM tool). The source code extends the piStar tool and provides a modeling and analysing environment in the web for GODA. The pistarGODA modeling and analyzing environment is available online at Heroku.

Uncertainty

The artifact allows the modeling of different sources and classes of uncertainty. We refer to our article to learn more about each of them.

• The first class of uncertainty considered, which is uncertainty related to the system itself, one can model the sources (1) future parameter value, and (2) incompleteness. Future parameter value is represented by realibility and cost parameters of a goal model leaf-tasks. For now, the reliability is set as a fixed parameter for each leaf-task, whose value changes at runtime. The cost parameters can be modeled as a weight function. Incompleteness is represented by a generic node in the goal model, as specified here.
• The second class of uncertainty is uncertainty related to system goals, in which we focus on the specification of goals (concerning the ability of accurately specific stakeholders' preferences). To model it, we specify a frequency of use parameter for each leaf-task in the goal model. For now, such parameter in intrinsicly assoaciated to each leaf-task and automatically generated by the artifact.
• The last class of uncertainty considered is uncertainty related to the environment. In this class, one can model (1) execution context, and (2) noise in sensing. Execution context is modeled as context conditions (see here). To support noise in sensing, one should follow the modeling guideline for systems when using sensors, in which a sensor's data collection behavior is decomposed in three essential sub-tasks: read data, filter data and transfer data (for more information about each sub-task, we refer to Section III.A of our article.

Modeling Instructions

Modeling Menu

Clear

Clear the current goal model.

Load model

Load existent .txt file. Folder examples contains examples of goal models to be loaded.

Save file

Save the model as a .txt file for future use.

Parametric formula generation

Once the goal model is finished, generate its reliability and cost parametric formulas (result.out and cost.out, respectively). Other outputs are: MDP model (file .nm) and PCTL properties (files .pctl) to be evaluated by a PRISM model checker; and Script eval_fomula.sh to evaluate the verification time of the formulae. We refer to Section III.C of our article to understand the practical use of the formulae.

Generate PRISM MDP code

Once the goal model is finished, generate its correspondent MDP model in PRISM language (file .nm).

New model

Use the upper menu to insert entities (actor, goals, tasks) and their refinements (AND/OR-Decomposition) into your new goal model. First, insert a new actor. Then, you can create goals/tasks inside your new actor. Double click on the entity to edit its label. We refer to our article's Background section for further knowledge on goal modeling.

Adding properties

Your root goal should always have the "selected true" property:

• Click on the root goal > Add property > Name: selected > Value: true

To add a context condition in a goal:

• Click on the specific goal > Add property > Name: creationProperty > Value: assertion condition context

To add a context condition in a task:

• Click on the specific task > Add property > Name: creationProperty > Value: assertion trigger context

The context follows a specific regex.

Context condition regex

Context conditions are formed by a proposition of context facts. You must use valid PRISM operators to define your context facts and to compose a valid context annotation. Valid operators are:

<, <=, >, >=, =, !=, &, |

Operands are the context facts and the values they can be compared with. Context facts are represented as meta-variables. They must start with normal, non-digit characters and may end with digits. Values can be either boolean (true or false), integer or double.

Valid context annotation examples are:

• BATTERY >= 50
• CONNECTION = true
• BATTERY >= 50 & CONNECTION = true

Syntax used to label Goals and Tasks

Goals and tasks must have a lable according to the rules bellow. The general form of a lable is:

• IDENTIFIER: DESCRIPTION [decision-making annotation]

IDENTIFIER: DESCRIPTION

• Goals must use a prefix G followed by an unique numeric ID within the CRGM Goals set followed by a colon followed by a textual description . Ex: G1: Goal description, G2: Goal description, G3: Goal description

• Same rule used for Goals, except that:

  • Tasks must use a prefix T
  • Task level refers to the depth level after the goal they succeed
  • Level 1 tasks' prefix should be followed by an unique numeric ID within their context. Ex:
    • T1: Task description (for goal G2)
    • T1: Task description (for goal G3)
  • Second and subsequent level tasks prefix must be followed by the same unique numeric ID of its first level task, a dot and an unique numeric ID. Ex:
    • T1.1: Tsk Dsc, T1.2: Tsk dsc (descendants of T1)
    • T1.11: Tsk dsc, T1.12: Tsk dsc (descendantas of T1 -> T1.1)
    • T1.111: Tsk dsc, T1.112: Tsk dsc, T1.113: Tsk dsc... (descendants of T1 -> T1.11) and so forward.

[Decision-making annotation]

• Any non-leaf node (goal or task) that is refined/decomposed into two or more sub-nodes can have a decision-making annotation as part of its label. A node with this annotation can be satisfied by any combination of its specified children.
• decision-making annotations must be inside brakets and be placed after DESCRIPTION. Ex:
• G1:Goal description [DM(G2,G3,G4)], meaning that the fulfillment of any combination of G2, G3, G4 will result in the fulfillment of G1.
• decision-making annotations must be separated from DESCRIPTION with space(s). As long as it is inside the brakets and after DESCRIPTION, it will be parsed.

Incomplete Node

• Incomplete node represents a goal/task that is not entirely known at design-time, leaving the goal model incomplete. In the goal model (and subsequent parametric formula and MDP model), it works as a placeholder for a runtime goal/task.
• The label follows the syntax used to label goals and tasks, but the final unique numeric ID is replaced by X. Ex: TX, T1.X, T1.1X.
• Such node cannot have children nodes.

Weigth function

• Any leaf-task can be annotated with a weigth function that determines the cost/reward of executing such node.
• The weight function regex is: W = value, in which value can be either a double, a string variable, or both.
• The weight function must be inside brackets and be placed after DESCRIPTION. Ex: T1.1: Leaf-task description [W = 0.1x]
• The weight function must be separated from DESCRIPTION with space(s). As long as it is inside the brakets and after DESCRIPTION, it will be parsed.

Example

Body Sensor Network contextual goal model augmented with uncertainty used as a case study in our article. For more information, we refer to the article's Background section.

Development

Installation

To run the source code locally:

• Clone this repo: $ git clone https://github.com/gabrielasolano/Seams2019.git
• Import it as a Maven project in your IDE.
• Run as java application, selecting Application from br.unb.cic.integration.
• In your browser: localhost:8080 to use the framework.

These are the steps when using the Eclipse IDE. Adaptation may be needed when using a different IDE.

Updating an ANTLR grammar

• After updating any .g4 grammar file (for context or runtime annotations), use maven to recompile the java classes for your language (regex, parser, etc):
  • At the project root folder, use (linux): mvn install -Dmaven.test.skip=true
  • At the project root folder, use (linux): mvn package -e -Dmaven.test.skip=true
  • Refresh the project folder in your IDE (Eclipse, etc.)

Bugs? Doubts?

• Look for existing issues or create a new one describing your problem or doubt
• Contact the author by email
  • felix.solano[at]gmail[dot]com
  • TODO: add other team members contacts here