MDE: Modelling the DEVS formalism

Transcription

1 MDE: Modelling the DEVS formalism Yentl Van Tendeloo January 24, / 25

2 The problem: Difficult to comprehend c l a s s Root ( CoupledDEVS ) : d e f i n i t ( s e l f ) : CoupledDEVS. i n i t ( s e l f ) s e l f. gen = s e l f. addsubmodel ( G e n e r a t o r ( ) ) s e l f. seg1 = s e l f. addsubmodel ( RoadSegment ( ) ) s e l f. seg2 = s e l f. addsubmodel ( RoadSegment ( ) ) s e l f. seg3 = s e l f. addsubmodel ( RoadSegment ( ) ) s e l f. c o l = s e l f. addsubmodel ( C o l l e c t o r ( ) ) s e l f. c o n n e c t P o r t s ( s e l f. gen. Q send, s e l f. seg1. Q recv ) s e l f. c o n n e c t P o r t s ( s e l f. gen. c a r o u t, s e l f. seg1. c a r i n ) s e l f. c o n n e c t P o r t s ( s e l f. seg1. Q sack, s e l f. gen. Q rack ) s e l f. c o n n e c t P o r t s ( s e l f. seg1. Q send, s e l f. seg2. Q recv ) s e l f. c o n n e c t P o r t s ( s e l f. seg1. c a r o u t, s e l f. seg2. c a r i n ) s e l f. c o n n e c t P o r t s ( s e l f. seg2. Q sack, s e l f. seg1. Q rack ) s e l f. c o n n e c t P o r t s ( s e l f. seg2. Q send, s e l f. seg3. Q recv ) s e l f. c o n n e c t P o r t s ( s e l f. seg2. c a r o u t, s e l f. seg3. c a r i n ) s e l f. c o n n e c t P o r t s ( s e l f. seg3. Q sack, s e l f. seg2. Q rack ) s e l f. c o n n e c t P o r t s ( s e l f. seg3. c a r o u t, s e l f. c o l. c a r i n ) 2 / 25

3 The problem: Difficult to comprehend c l a s s Root ( CoupledDEVS ) : d e f i n i t ( s e l f ) : CoupledDEVS. i n i t ( s e l f ) s e l f. gen = s e l f. addsubmodel ( G e n e r a t o r ( ) ) s e l f. seg1 = s e l f. addsubmodel ( RoadSegment ( ) ) s e l f. seg2 = s e l f. addsubmodel ( RoadSegment ( ) ) s e l f. seg3 = s e l f. addsubmodel ( RoadSegment ( ) ) s e l f. c o l = s e l f. addsubmodel ( C o l l e c t o r ( ) ) s e l f. c o n n e c t P o r t s ( s e l f. gen. Q send, s e l f. seg1. Q recv ) s e l f. c o n n e c t P o r t s ( s e l f. gen. c a r o u t, s e l f. seg1. c a r i n ) s e l f. c o n n e c t P o r t s ( s e l f. seg1. Q sack, s e l f. gen. Q rack ) s e l f. c o n n e c t P o r t s ( s e l f. seg1. Q send, s e l f. seg2. Q recv ) s e l f. c o n n e c t P o r t s ( s e l f. seg1. c a r o u t, s e l f. seg2. c a r i n ) s e l f. c o n n e c t P o r t s ( s e l f. seg2. Q sack, s e l f. seg1. Q rack ) s e l f. c o n n e c t P o r t s ( s e l f. seg2. Q send, s e l f. seg3. Q recv ) s e l f. c o n n e c t P o r t s ( s e l f. seg2. c a r o u t, s e l f. seg3. c a r i n ) s e l f. c o n n e c t P o r t s ( s e l f. seg3. Q sack, s e l f. seg2. Q rack ) s e l f. c o n n e c t P o r t s ( s e l f. seg3. c a r o u t, s e l f. c o l. c a r i n ) Solution: Graphical representation 2 / 25

4 The problem: Different simulators i n t main ( i n t, c h a r a r g v ){ adevs : : Digraph<Event > dig ; Generator gen = new Generator ( ) ; RoadSegment seg1 = new RoadSegment ( ) ; RoadSegment seg2 = new RoadSegment ( ) ; RoadSegment seg3 = new RoadSegment ( ) ; C o l l e c t o r c o l = new C o l l e c t o r ( ) ; d i g. add ( gen ) ; d i g. add ( seg1 ) ; d i g. add ( seg2 ) ; d i g. add ( seg3 ) ; d i g. add ( c o l ) ; dig. couple ( gen, gen >Q send, seg1, seg1 >Q recv ) ; d i g. c o u p l e ( gen, gen >c a r o u t, seg1, seg1 >c a r i n ) ; dig. couple ( seg1, seg1 >Q sack, gen, gen >Q rack ) ; dig. couple ( seg1, seg1 >Q send, seg2, seg2 >Q recv ) ; d i g. c o u p l e ( seg1, seg1 >c a r o u t, seg2, seg2 >c a r i n ) ; dig. couple ( seg2, seg2 >Q sack, seg1, seg1 >Q rack ) ; dig. couple ( seg2, seg2 >Q send, seg3, seg3 >Q recv ) ; d i g. c o u p l e ( seg2, seg2 >c a r o u t, seg3, seg3 >c a r i n ) ; dig. couple ( seg3, seg3 >Q sack, seg2, seg2 >Q rack ) ; d i g. c o u p l e ( seg3, seg3 >c a r o u t, c o l, c o l >c a r i n ) ; 3 / 25

5 The problem: Different simulators i n t main ( i n t, c h a r a r g v ){ adevs : : Digraph<Event > dig ; Generator gen = new Generator ( ) ; RoadSegment seg1 = new RoadSegment ( ) ; RoadSegment seg2 = new RoadSegment ( ) ; RoadSegment seg3 = new RoadSegment ( ) ; C o l l e c t o r c o l = new C o l l e c t o r ( ) ; d i g. add ( gen ) ; d i g. add ( seg1 ) ; d i g. add ( seg2 ) ; d i g. add ( seg3 ) ; d i g. add ( c o l ) ; dig. couple ( gen, gen >Q send, seg1, seg1 >Q recv ) ; d i g. c o u p l e ( gen, gen >c a r o u t, seg1, seg1 >c a r i n ) ; dig. couple ( seg1, seg1 >Q sack, gen, gen >Q rack ) ; dig. couple ( seg1, seg1 >Q send, seg2, seg2 >Q recv ) ; d i g. c o u p l e ( seg1, seg1 >c a r o u t, seg2, seg2 >c a r i n ) ; dig. couple ( seg2, seg2 >Q sack, seg1, seg1 >Q rack ) ; dig. couple ( seg2, seg2 >Q send, seg3, seg3 >Q recv ) ; d i g. c o u p l e ( seg2, seg2 >c a r o u t, seg3, seg3 >c a r i n ) ; dig. couple ( seg3, seg3 >Q sack, seg2, seg2 >Q rack ) ; d i g. c o u p l e ( seg3, seg3 >c a r o u t, c o l, c o l >c a r i n ) ; Solution: Independent language 3 / 25

6 Global overview Source: Hongyan Song. Infrastructure for devs modelling and experimentation. Master s thesis, McGill University, / 25

7 Overview: Modelling In AToM 3 Compliant to a metamodel Should have a code generation feature For intermediate language (Modelica) Automatically generated environment! 5 / 25

8 Intermediate Language: Modelica Modelling language Relatively mature Language features for both continuous and discrete models Standard library Re-usability 6 / 25

9 Modelica Example c l a s s G e n e r a t o r extends AtomicDEVS ; output DevsPort p o u t ; G e n e r a t o r S t a t e s t a t e ( ) ; f u n c t i o n timeadvance output I n t e g e r t i m e s p a n ; a l g o r i t h m... end timeadvance ; end G e n e r a t o r ; c l a s s Root extends CoupledDEVS ; output DevsPort p o u t ; P r o c e s s o r i n s 2 ( ) ; e q u a t i o n c o n n e c t ( i n s 2. p o u t, p o u t ) ; end Root ; 7 / 25

10 Overview: Verification Use the µmodelica compiler 1 Statically check the model No accesses to inaccessible variables Possibly perform optimisations Currently (nearly) no verification happens 1 Weigao Xu. The design and implementation of the modelica compiler. Master s thesis, McGill University, / 25

11 Overview: Simulation Use the PythonDEVS simulator 2 Save the trace file Text output XML output VCD output 2 Jean-Sébastien Bolduc and Hans Vangheluwe. The modelling and simulation package pythondevs for classical hierarchical devs. Technical report, MSDL Technical Report, / 25

12 Overview: Validation Compare the output of the simulation to the specifications Textual simulation traces are unclear Visualize! 10 / 25

13 Overview: Validation (example) Source: Hongyan Song. Infrastructure for devs modelling and experimentation. Master s thesis, McGill University, / 25

14 Overview: Features Model reuse At AToM 3 level At Modelica level At PythonDEVS level Clear boundaries Open structure 12 / 25

15 My contribution Modelling Updating the existing metamodel Include experiment data Verification Update the µmodelica compiler Introduce a flattening phase 13 / 25

16 Modelling in AToM 3 : metamodel hasstate atomicdevsv2 containmodelv2 Actions: > connect > disconnect Attributes: coupleddevsv2 - name :: String Attributes: - isvisible :: Boolean - name :: String Actions: > connect < disconnect DevsExperiment Attributes: - experimentname :: String - model :: String - classvariables :: List - isvisible :: Boolean - package :: String DevsState Attributes: - name :: String - attributes :: List - parameters :: List - attributes :: List - init :: Text - exttransition :: Text - inttransition :: Text - outputfunc :: Text - timeadvance :: Text - classvariables :: List - parameters :: List - attributes :: List - init :: Text - ismainmodel :: Boolean insdevsv2 Attributes: - name :: String - type :: String - parameters :: Text - isvisible :: Boolean - parameters :: Text - end_time :: Float - verbose :: Boolean - trace :: Boolean containsportv2 containsstatev2 Actions: portdevsv2 internaltransitionv2 Actions: > connect Attributes: Attributes: > connect < disconnect - name :: String - condition :: Text < disconnect - porttype :: Enum - action :: Text eventdevsv2 externaltransitionv2 Attributes: - condition :: Text - action :: Text statedevsv2 Attributes: - name :: String - initial :: Boolean - timeadvance :: Text - output :: Text - extaction :: Text - intaction :: Text Attributes: - name :: String - classvariables :: List channelv2 - parameters :: List Actions: - attributes :: List > connect - init :: Text < disconnect - str :: Text < checkvalidity - otherfunc :: Text 14 / 25

17 Modelling in AToM 3 : example model Generator generating p_out idle p_out Processor processing Experiment: Model: Root Experiment TimeAdvance: timespan := 1; Output:evt := Job(0.3); poke(p_out, evt); p_in TimeAdvance: timespan := INFINITY; Output: TimeAdvance: timespan := event.jobsize; Output: poke(p_out, event); Package: Parameters: End time: Verbose: True Trace: False Root Job Parameters: p_out ins_7 : Generator ins_6 p_in CoupledProcessor : p_out p_in ins_2 : Processor p_out p_out jobsize type=float init.value=3.5 Attributes: CoupledProcessor p_in ins_3 p_in : Processor p_out ins_4 : Processor p_out p_in p_out 15 / 25

18 DEVS: Messages 16 / 25

19 Verification: Flattening What is flattening? Perform the closure under coupling on a real model Together with direct connection Use of flattening Increased performance The proof was right! 17 / 25

20 Verification: Direct connection What is direct connection? Remove the complete hierarchy Just 1 coupled model Connections become one-step Problems! 3 Z functions need to be rewritten select functions need to be rewritten 3 Bin Chen and Hans Vangheluwe. Symbolic flattening of devs models. In 2010 Summer Simulation Multiconference, SummerSim 10, pages , San Diego, CA, USA, Society for Computer Simulation International 18 / 25

21 Verification: Direct connection 19 / 25

22 Verification: Compilation Still need to compile the generated modelica code Some slight changes were done here 20 / 25

23 Evaluation Simulation time of Wide models using the Old PyDEVS 70 Flattened Quadratic Non-Flattened 60 Quadratic Simulation time of Wide models using the New PyDEVS 40 Flattened Quadratic 35 Non-Flattened Linear Time (s) Time (s) Width Width 21 / 25

24 Evaluation Simulation time of Deep models using the Old PyDEVS 11 Flattened 10 Quadratic Non-Flattened Constant 9 Simulation time of Deep models using the New PyDEVS 0.33 Flattened Linear 0.32 Non-Flattened Constant Time (s) 6 5 Time (s) Depth Depth 22 / 25

25 Evaluation: Gain These optimisations are simulator-independent! Optimisations will propagate Difference depends on the simulator This comparison was with an optimised simulator 4 4 Yentl Van Tendeloo. Research internship 1: Optimizing pydevs. Technical report, MSDL, / 25

26 Future work Different dimensions Optimise flattening Implement Verifications Implement other back-ends Some details were ignored Z and select function 24 / 25

27 Demo Time for a demo! 25 / 25

28 Jean-Sébastien Bolduc and Hans Vangheluwe. The modelling and simulation package pythondevs for classical hierarchical devs. Technical report, MSDL Technical Report, Bin Chen and Hans Vangheluwe. Symbolic flattening of devs models. In 2010 Summer Simulation Multiconference, SummerSim 10, pages , San Diego, CA, USA, Society for Computer Simulation International. Hongyan Song. Infrastructure for devs modelling and experimentation. Master s thesis, McGill University, Yentl Van Tendeloo. Research internship 1: Optimizing pydevs. Technical report, MSDL, Weigao Xu. The design and implementation of the modelica compiler. 25 / 25

29 Master s thesis, McGill University, / 25