[PPT] - ON COMBINING TRIPLE GRAPH GRAMMARS AND LINEAR OPTIMISATION PowerPoint Presentation

SLIDE 1

ON COMBINING TRIPLE GRAPH GRAMMARS AND LINEAR OPTIMISATION TECHNIQUES

Erhan Leblebici, Anthony Anjorin, Andy Schürr

SLIDE 2

A Real-World Example: Overview

2

Robin Oppermann:  A Configurable, Model-Driven Approach to Optimal Scheduling using Triple Graph Grammars and Linear Progamming. Ongoing Master’s Thesis, Paderborn University in collaboration with dSpace

SLIDE 3

A Real-World Example: Overview

2

Robin Oppermann:  A Configurable, Model-Driven Approach to Optimal Scheduling using Triple Graph Grammars and Linear Progamming. Ongoing Master’s Thesis, Paderborn University in collaboration with dSpace

A series of recurring manual tests have to be executed.

SLIDE 4

A Real-World Example: Overview

2

Robin Oppermann:  A Configurable, Model-Driven Approach to Optimal Scheduling using Triple Graph Grammars and Linear Progamming. Ongoing Master’s Thesis, Paderborn University in collaboration with dSpace

A series of recurring manual tests have to be executed. A test manager has to create a test schedule assigning test items to developers.

SLIDE 5

A Real-World Example: Overview

2

Robin Oppermann:  A Configurable, Model-Driven Approach to Optimal Scheduling using Triple Graph Grammars and Linear Progamming. Ongoing Master’s Thesis, Paderborn University in collaboration with dSpace

A series of recurring manual tests have to be executed. A test manager has to create a test schedule assigning test items to developers. Developers are only available for this purpose a few hours a week due to other responsibilities.

SLIDE 6

A Real-World Example: Overview

2

Robin Oppermann:  A Configurable, Model-Driven Approach to Optimal Scheduling using Triple Graph Grammars and Linear Progamming. Ongoing Master’s Thesis, Paderborn University in collaboration with dSpace

A series of recurring manual tests have to be executed. A test manager has to create a test schedule assigning test items to developers. Developers are only available for this purpose a few hours a week due to other responsibilities. Developers are also on vacation now and then…

SLIDE 7

A Real-World Example: Overview

2

Robin Oppermann:  A Configurable, Model-Driven Approach to Optimal Scheduling using Triple Graph Grammars and Linear Progamming. Ongoing Master’s Thesis, Paderborn University in collaboration with dSpace

A series of recurring manual tests have to be executed. A test manager has to create a test schedule assigning test items to developers. Developers are only available for this purpose a few hours a week due to other responsibilities. Developers are also on vacation now and then… … and might not have the required expertise for all tasks.

SLIDE 8

A Real-World Example: Metamodels

3

duration: EInt Execution Area Task Responsibility Person hours: EInt week: WEEKS Availability

SLIDE 9

A Real-World Example: Metamodels

3

duration: EInt Execution Area Task Responsibility Person hours: EInt week: WEEKS Availability

There are predefined areas and people in charge of these areas.

SLIDE 10

A Real-World Example: Metamodels

3

duration: EInt Execution Area Task Responsibility Person hours: EInt week: WEEKS Availability

There are predefined areas and people in charge of these areas. A test schedule maps executions to availabilities.

SLIDE 11

A Real-World Example: Allocation Rules

4

t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

SLIDE 12

A Real-World Example: Allocation Rules

4

t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

In general, anyone can do anything as long as they have the time for it…

SLIDE 13

A Real-World Example: Allocation Rules

5

t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

ar:Area r:Responsibility p’:Person

SLIDE 14

A Real-World Example: Allocation Rules

5

t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

ar:Area r:Responsibility p’:Person

Only test things that are not in your area of responsibility?

SLIDE 15

A Real-World Example: Allocation Rules

6

t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

ar:Area r:Responsibility p’:Person

t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

ar:Area r:Responsibility

SLIDE 16

A Real-World Example: Allocation Rules

6

t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

ar:Area r:Responsibility p’:Person

t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

ar:Area r:Responsibility

Or perhaps being an expert makes you the best tester possible?

SLIDE 17

A Real-World Example: Allocation Rules

7

t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

ar:Area r:Responsibility p’:Person t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

ar:Area r:Responsibility

t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

e’:Execution a’:Availability

SLIDE 18

A Real-World Example: Allocation Rules

7

t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

ar:Area r:Responsibility p’:Person t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

ar:Area r:Responsibility

t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

e’:Execution a’:Availability

Should the same person test as many executions of the same task as possible? Or is this a terrible idea?

SLIDE 19

A Real-World Example: Allocation Rules

8

t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

ar:Area r:Responsibility p’:Person

t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

ar:Area r:Responsibility t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

e’:Execution a’:Availability

SLIDE 20

Similar Application Domains

9

SLIDE 21

Similar Application Domains

9

1. Allocation Engineering:
Tasks to resources
Programs to ECUs
Functions to nodes in a network
…

SLIDE 22

Similar Application Domains

9

1. Allocation Engineering:
Tasks to resources
Programs to ECUs
Functions to nodes in a network
…
2. Traceability Maintenance:
Suggest traceability links
Check manually created traceability links
Flag “suspect links” after changes

SLIDE 23

Similar Application Domains

9

1. Allocation Engineering:
Tasks to resources
Programs to ECUs
Functions to nodes in a network
…
2. Traceability Maintenance:
Suggest traceability links
Check manually created traceability links
Flag “suspect links” after changes
3. Model Synchronisation:
Start with existing, independently

created models

Identify inconsistencies

SLIDE 24

Our Approach

10

Erhan Leblebici, Anthony Anjorin, Andy Schürr:  Inter-model Consistency Checking Using Triple Graph Grammars and Linear Optimization Techniques. FASE 2017: 191-207 Erhan Leblebici:  Inter-Model Consistency Checking and Restoration with Triple Graph Grammars. PhD Thesis, Darmstadt University of Technology, Germany 2018

2017: 2018:

Nils Weidmann:  Consistency Management via a Combination of Triple Graph Grammars and Integer Linear Programming. Master’s Thesis, Paderborn University, Germany 2018

2018:

Erhan Leblebici:  Towards a Graph Grammar-Based Approach to Inter-Model Consistency Checks with Traceability Support. Bx@ETAPS 2016: 35-39

2016:

SLIDE 25

Our Approach

10

Erhan Leblebici, Anthony Anjorin, Andy Schürr:  Inter-model Consistency Checking Using Triple Graph Grammars and Linear Optimization Techniques. FASE 2017: 191-207 Erhan Leblebici:  Inter-Model Consistency Checking and Restoration with Triple Graph Grammars. PhD Thesis, Darmstadt University of Technology, Germany 2018

2017: 2018:

Nils Weidmann:  Consistency Management via a Combination of Triple Graph Grammars and Integer Linear Programming. Master’s Thesis, Paderborn University, Germany 2018

2018:

Erhan Leblebici:  Towards a Graph Grammar-Based Approach to Inter-Model Consistency Checks with Traceability Support. Bx@ETAPS 2016: 35-39

2016:

SLIDE 26

Our Approach

10

Erhan Leblebici, Anthony Anjorin, Andy Schürr:  Inter-model Consistency Checking Using Triple Graph Grammars and Linear Optimization Techniques. FASE 2017: 191-207 Erhan Leblebici:  Inter-Model Consistency Checking and Restoration with Triple Graph Grammars. PhD Thesis, Darmstadt University of Technology, Germany 2018

2017: 2018:

Nils Weidmann:  Consistency Management via a Combination of Triple Graph Grammars and Integer Linear Programming. Master’s Thesis, Paderborn University, Germany 2018

2018:

Erhan Leblebici:  Towards a Graph Grammar-Based Approach to Inter-Model Consistency Checks with Traceability Support. Bx@ETAPS 2016: 35-39

2016:

Basic idea of how to perform consistency checking with TGGs

SLIDE 27

Our Approach

11

Erhan Leblebici, Anthony Anjorin, Andy Schürr:  Inter-model Consistency Checking Using Triple Graph Grammars and Linear Optimization Techniques. FASE 2017: 191-207 Erhan Leblebici:  Inter-Model Consistency Checking and Restoration with Triple Graph Grammars. PhD Thesis, Darmstadt University of Technology, Germany 2018

2017: 2018:

Nils Weidmann:  Consistency Management via a Combination of Triple Graph Grammars and Integer Linear Programming. Master’s Thesis, Paderborn University, Germany 2018

2018:

Erhan Leblebici:  Towards a Graph Grammar-Based Approach to Inter-Model Consistency Checks with Traceability Support. Bx@ETAPS 2016: 35-39

2016:

Full details, implementation, and evaluation in eMoflon

SLIDE 28

Our Approach

12

Erhan Leblebici, Anthony Anjorin, Andy Schürr:  Inter-model Consistency Checking Using Triple Graph Grammars and Linear Optimization Techniques. FASE 2017: 191-207 Erhan Leblebici:  Inter-Model Consistency Checking and Restoration with Triple Graph Grammars. PhD Thesis, Darmstadt University of Technology, Germany 2018

2017: 2018:

Nils Weidmann:  Consistency Management via a Combination of Triple Graph Grammars and Integer Linear Programming. Master’s Thesis, Paderborn University, Germany 2018

2018:

Erhan Leblebici:  Towards a Graph Grammar-Based Approach to Inter-Model Consistency Checks with Traceability Support. Bx@ETAPS 2016: 35-39

2016:

Remaining formal proofs, industrial case with Siemens

SLIDE 29

Our Approach

13

Erhan Leblebici, Anthony Anjorin, Andy Schürr:  Inter-model Consistency Checking Using Triple Graph Grammars and Linear Optimization Techniques. FASE 2017: 191-207 Erhan Leblebici:  Inter-Model Consistency Checking and Restoration with Triple Graph Grammars. PhD Thesis, Darmstadt University of Technology, Germany 2018

2017: 2018:

Nils Weidmann:  Consistency Management via a Combination of Triple Graph Grammars and Integer Linear Programming. Master’s Thesis, Paderborn University, Germany 2018

2018:

Erhan Leblebici:  Towards a Graph Grammar-Based Approach to Inter-Model Consistency Checks with Traceability Support. Bx@ETAPS 2016: 35-39

2016:

Generalisation of the approach to

ther consistency management

tasks (work in progress)

SLIDE 30

Step 1: Collect all Candidates

14

t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

ar:Area r:Responsibility p’:Person

t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

ar:Area r:Responsibility t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

e’:Execution a’:Availability

SLIDE 31

Step 1: Collect all Candidates

14

t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

ar:Area r:Responsibility p’:Person

t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

ar:Area r:Responsibility t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

e’:Execution a’:Availability

Use allocation rules (derived from a TGG) to create all possible correspondence links.

SLIDE 32

Step 1: Collect all Candidates

14

t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

ar:Area r:Responsibility p’:Person

t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

ar:Area r:Responsibility t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

e’:Execution a’:Availability

Use allocation rules (derived from a TGG) to create all possible correspondence links.

SLIDE 33

Step 1: Collect all Candidates

14

t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

ar:Area r:Responsibility p’:Person

t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

ar:Area r:Responsibility t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

e’:Execution a’:Availability

Use allocation rules (derived from a TGG) to create all possible correspondence links. While these links are

nly candidates, they

still make sense locally

SLIDE 34

Step 1: Collect all Candidates

14

t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

ar:Area r:Responsibility p’:Person

t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

ar:Area r:Responsibility t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

e’:Execution a’:Availability

Use allocation rules (derived from a TGG) to create all possible correspondence links. While these links are

nly candidates, they

still make sense locally This step requires a necessary and sufficient condition for termination!

SLIDE 35

Step 1: Collect all Candidates

14

t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

ar:Area r:Responsibility p’:Person

t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

ar:Area r:Responsibility t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

e’:Execution a’:Availability

Use allocation rules (derived from a TGG) to create all possible correspondence links. While these links are

nly candidates, they

still make sense locally This step exploits an incremental graph pattern matcher This step requires a necessary and sufficient condition for termination!

SLIDE 36

Step 2: Derive ILP

15

SLIDE 37

Step 2: Derive ILP

15

SLIDE 38

Step 2: Derive ILP

15

max ⃗ c ⋅ ⃗ x A ⃗ x ≤ ⃗ b

SLIDE 39

Step 2: Derive ILP

16

max ⃗ c ⋅ ⃗ x A ⃗ x ≤ ⃗ b

Which candidates will be part of the solution?

⃗ x ∈ ℤn

2

SLIDE 40

Step 2: Derive ILP

17

max ⃗ c ⋅ ⃗ x A ⃗ x ≤ ⃗ b

Domain-specific weights for each candidate

⃗ c ∈ ℝn

SLIDE 41

Step 2: Derive ILP

17

max ⃗ c ⋅ ⃗ x A ⃗ x ≤ ⃗ b

Domain-specific weights for each candidate

⃗ c ∈ ℝn

E.g., prefer assigning multiple executions of the same task to the same person

SLIDE 42

Step 2: Derive ILP

18

max ⃗ c ⋅ ⃗ x A ⃗ x ≤ ⃗ b

Constraints to ensure that the chosen solution is in the language of the TGG.

SLIDE 43

Step 2: Derive ILP

19

max ⃗ c ⋅ ⃗ x A ⃗ x ≤ ⃗ b

t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

e’:Execution a’:Availability

SLIDE 44

Step 2: Derive ILP

19

max ⃗ c ⋅ ⃗ x A ⃗ x ≤ ⃗ b

t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

e’:Execution a’:Availability

E.g, let’s assume all candidates for creating this link are:

x1, x2, x3

SLIDE 45

Step 2: Derive ILP

19

max ⃗ c ⋅ ⃗ x A ⃗ x ≤ ⃗ b

t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

e’:Execution a’:Availability

E.g, let’s assume all candidates for creating this link are:

x1, x2, x3

This candidate requires at least one of them to exist:

x4 ⇒ x1 ∨ x2 ∨ x3 x4 ≤ x1 + x2 + x3

SLIDE 46

Step 2: Derive ILP

19

max ⃗ c ⋅ ⃗ x A ⃗ x ≤ ⃗ b

t:Task p:Person e:Execution a:Availability

e.duration ≤ a.hours ++

e’:Execution a’:Availability

E.g, let’s assume all candidates for creating this link are:

x1, x2, x3

This candidate requires at least one of them to exist:

x4 ⇒ x1 ∨ x2 ∨ x3 x4 ≤ x1 + x2 + x3

All such inequalities are collected to form:

A ∈ ℝm×n, b ∈ ℝn

SLIDE 47

Step 3: Solve (Optimise) ILP and Interpret Solution

20

max ⃗ c ⋅ ⃗ x A ⃗ x ≤ ⃗ b

⃗ x *

SLIDE 48

Step 3: Solve (Optimise) ILP and Interpret Solution

20

max ⃗ c ⋅ ⃗ x A ⃗ x ≤ ⃗ b

⃗ x *

This step exploits mature ILP solvers

SLIDE 49

Step 3: Solve (Optimise) ILP and Interpret Solution

21

⃗ x *

SLIDE 50

Step 3: Solve (Optimise) ILP and Interpret Solution

21

⃗ x *

SLIDE 51

Step 3: Solve (Optimise) ILP and Interpret Solution

21

⃗ x *

SLIDE 52

Step 3: Solve (Optimise) ILP and Interpret Solution

21

⃗ x *

Our approach is tolerant in the sense that we can determine partial solutions (all variables are set to 0 in the worst case)

SLIDE 53

Ongoing and Future Work

22

Operation Source Corr Target CC mark create mark CO mark mark mark FWD_OPT mark create create BWD_OPT create create mark

SLIDE 54

Ongoing and Future Work

22

Operation Source Corr Target CC mark create mark CO mark mark mark FWD_OPT mark create create BWD_OPT create create mark

Our initial focus (Consistency Check via correspondence link creation)

SLIDE 55

Ongoing and Future Work

23

Operation Source Corr Target CC mark create mark CO mark mark mark FWD_OPT mark create create BWD_OPT create create mark

Check Only: Check existing triple for consistency

SLIDE 56

Ongoing and Future Work

24

Operation Source Corr Target CC mark create mark CO mark mark mark FWD_OPT mark create create BWD_OPT create create mark

Normal initial (batch) fwd and bwd transformations; but now complete, tolerant, and optimal wrt. to an objective function

SLIDE 57

Ongoing and Future Work

25

Operation Source Corr Target CC mark create mark CO mark mark mark FWD_OPT mark create create BWD_OPT create create mark

All definitions, proofs, and most parts of the implementation can be formulated generically and configured for each case using the entries in this table!

SLIDE 58

Try it out! www.emoflon.org

26