Models for Efficient Timed Verification

Transcription

1 Models for Efficient Timed Verification François Laroussinie LSV / ENS de Cachan CNRS UMR 8643 Monterey Workshop - Composition of embedded systems

2 Model checking System Properties Formalizing step? ϕ

3 Model checking We want... Expressive models: communication, variables, timing constraints, probabilities... Expressive specification languages: natural, powerful, intuitive...

4 Model checking We want... Expressive models: communication, variables, timing constraints, probabilities... Expressive specification languages: natural, powerful, intuitive... AND EFFICIENT ALGORITHM! Main limit of model checking = state explosion problem

5 State explosion problem An example: a protocol...

6 State explosion problem An example: a protocol a sender

7 State explosion problem An example: a protocol a buffer 4 a sender

8 State explosion problem An example: a protocol a buffer 4 a sender 1 2 a receiver

9 State explosion problem An example: a protocol a sender a buffer 1 a server 2 2 a receiver

10 State explosion problem An example: a protocol a sender a buffer 1 a server 2 2 a receiver

11 State explosion problem An example: a protocol a buffer 3 4 a sender 1 1 a server a receiver

12 State explosion problem

13 State explosion problem This has motivated symbolic methods heuristics: on-the-fly algorithms acceleration partial order... abstraction... Model checkers exist and they work rather nicely!

14 Verification of compositions of embedded systems Two difficulties: the composition... Usually we need to address verification of quantitative aspects: timing constraints (Real-time systems) probabilities costs data...

15 Verification of compositions of embedded systems Two difficulties: the composition... Usually we need to address verification of quantitative aspects: timing constraints (Real-time systems) probabilities costs data... Each one may induce a complexity blow-up. This can be measured by the structural complexity of verification problems: for ex. the composition and the state explosion problem.

16 State explosion problem This state explosion holds for synchronized systems, systems with boolean variables, 1-safe Petri nets etc. These are non-flat systems. In practice, a model S is described as a non-flat system (whose operational semantics is a flat transition system T)

17 State explosion problem This state explosion holds for synchronized systems, systems with boolean variables, 1-safe Petri nets etc. These are non-flat systems. In practice, a model S is described as a non-flat system (whose operational semantics is a flat transition system T) Complexity of model checking non-flat systems Model checking T = Φ non-flat syst. = Φ Reachability NLOGSPACE-C PSPACE-C CTL model checking P-C PSPACE-C AF µ-calculus P-C EXPTIME-C (Papadimitriou, Vardi, Kupferman, Wolper, Rabinovich,...)

18 Challenge Define new formalisms to model and verify the embedded systems, with... timing constraints probabilities costs, dynamic variables... extended specification languages

19 Challenge Define new formalisms to model and verify the embedded systems, with... timing constraints probabilities costs, dynamic variables... extended specification languages request AF grant request AF <200 grant request P >0.9 F <200 grant request Agt 1 P >0.9 F <200 grant

20 Challenge Define new formalisms to model and verify the embedded systems, with... timing constraints probabilities costs, dynamic variables... extended specification languages request AF grant request AF <200 grant request P >0.9 F <200 grant request Agt 1 P >0.9 F <200 grant without increasing (too much) the complexity!

21 Efficient timed model-checking obj: find timed models and timed specification languages for which verification can be done efficiently. To model simply timed systems To use to abstract complex timed systems To combine with other quantitative extensions

22 Outline 1 Timed specification languages 2 Timed models Discrete-time models Timed automata 3 Conclusion

23 Outline 1 Timed specification languages 2 Timed models Discrete-time models Timed automata 3 Conclusion

24 Classical verification problems Reachability of a control state S S? : (bi)simulation, etc. L(S) L(S)? : language inclusion (S A T ) + reachability : testing automata S = Φ with Φ a temporal logic formula : model checking

25 Timed properties Temporal properties: Any problem is followed by an alarm With CTL: ( ) AG problem AF alarm

26 Timed properties Temporal properties: Any problem is followed by an alarm With CTL: ( ) AG problem AF alarm Timed properties: Any problem is followed by an alarm in at most 10 time units

27 Timed properties Temporal properties: Any problem is followed by an alarm With CTL: ( ) AG problem AF alarm Timed properties: Any problem is followed by an alarm in at most 10 time units With TCTL: ( ) AG problem AF 10 alarm

28 Timed CTL We use TCTL whose formulae are built from: atomic propositions in Prop (For ex. Alarm, Problem,...) boolean combinators (,, ), and temporal operators tagged with timing constraints: E U c and A U c with {<,, =,, >} and c N. + all the standard abbreviations: AG c, AF c etc.

29 Timed CTL We use TCTL whose formulae are built from: atomic propositions in Prop (For ex. Alarm, Problem,...) boolean combinators (,, ), and temporal operators tagged with timing constraints: E U c and A U c with {<,, =,, >} and c N. + all the standard abbreviations: AG c, AF c etc. s = EϕU c ψ ρ = ρ ρ Exec(s) with s ρ s and Time(ρ ) c and s = ψ, and s = ϕ for all s < ρ s < ρ s

30 Outline 1 Timed specification languages 2 Timed models Discrete-time models Timed automata 3 Conclusion

31 Questions Studying timing constraints & complexity Two problems: 1 Are there simpler models with discrete time for which model checking can be efficient? 2 Are there classes of Timed Automata which allow efficient model checking?

32 Timed transition systems = a transition system + a notion of time used to define the semantics of real-time systems Let T be a time domain: N, R + or Q +. Timed transition system T = S, s 0,, l S is an (possibly infinite) set of states, s 0 S S T S l : S 2 Prop : assigns atomic propositions to states

33 Timed transition systems = a transition system + a notion of time used to define the semantics of real-time systems Let T be a time domain: N, R + or Q +. Timed transition system T = S, s 0,, l S is an (possibly infinite) set of states, s 0 S S T S l : S 2 Prop : assigns atomic propositions to states Every finite run σ in T has a finite duration Time(σ).

34 Outline 1 Timed specification languages 2 Timed models Discrete-time models Timed automata 3 Conclusion

35 Simple models for timed verification It is possible to use classical Kripke structures as timed models. There is no inherent concept of time: time elapsing is encoded by events. For example: each transition = one time unit (Emerson et al.) or: a tick proposition labels states where one t.u. elapses. Timed model checking can be efficient (polynomial-time)! Durational Transition Graphs extend these models without loosing efficiency.

36 Durational Transition Graph [0, ) Init state [7,45] New Card 0 Wait for Pincode. Read code [25,50] [25,50] 0 [0, ) [10;15] [0,366] Code OK Bad code Money+card [50,110] [0,7] Ask amount [5,10] Return card 10

37 Durational Transition Graph T = N A durational transition graph S is Q, R, l where Q is a (finite) set of states, R Q I Q is a total transition relation with duration l : Q 2 Prop labels every state with a subset of Prop. I = set of intervals [n, m] or [n, ) (with n, m N) q [n,m] q : moving from q to q takes some duration d in [n, m]. Two semantics are possible...

38 Two semantics for DTGs Consider the run: Init State 15 New Card 0 Wait for Pin C

39 Two semantics for DTGs Consider the run: Init State 15 New Card 0 Wait for Pin C Jump semantics: date: 0 Init State date: 15 New Card no intermediary states!... Simple semantics, no extra states. Not always natural. Difficult to synchronize two DTGs.

40 Two semantics for DTGs Consider the run: Init State 15 New Card 0 Wait for Pin C Jump semantics: date: 0 Init State date: 15 New Card no intermediary states!... Simple semantics, no extra states. Not always natural. Difficult to synchronize two DTGs. Continuous semantics: date: Init state date: 15 New card date: Wait for Pin C...

41 jump vs continuous semantics [2;4] r DTG S q [0;0] [3;3] s [1;2]

42 jump vs continuous semantics [2;4] r DTG S q [0;0] [3;3] s [1;2] jump sem. of S: 2 r 3 q s 1 2

43 jump vs continuous semantics [2;4] r DTG S q [0;0] [3;3] s [1;2] jump sem. of S: 2 r continuous sem. of S: 1 r, q 4 0 q, 0 1 q, 1 q, 2 q, s 1 s, 0 1 s, (q, i) : i time units have already been spent in q

44 Outline 1 Timed specification languages 2 Timed models Discrete-time models Timed automata 3 Conclusion

45 Timed Automata Example A = an automaton (locations and transitions) + clocks true, b, x := 0 x < 10, b, x := 0 OFF ON x 10 x = 10, i,

46 Timed automata - definition [Alur & Dill] A = an automaton (locations and transitions) + clocks Clocks progress synchronously with time (Time domain = R + ) Transitions: l g,a,r l T with: g is the guard, a is the label, r is the set of clocks to be reset to 0

47 Timed automata - definition [Alur & Dill] A = an automaton (locations and transitions) + clocks Clocks progress synchronously with time (Time domain = R + ) Transitions: l g,a,r l T with: g is the guard, a is the label, r is the set of clocks to be reset to 0 Semantics: States: (l, v) where v is a valuation for the clocks

48 Timed automata - definition [Alur & Dill] A = an automaton (locations and transitions) + clocks Clocks progress synchronously with time (Time domain = R + ) Transitions: l g,a,r l T with: g is the guard, a is the label, r is the set of clocks to be reset to 0 Semantics: States: (l, v) where v is a valuation for the clocks Action transition: { (l, v) 0 g,a,r a (l, v l l ) iff A, v = g, v = v[r 0]

49 Timed automata - definition [Alur & Dill] A = an automaton (locations and transitions) + clocks Clocks progress synchronously with time (Time domain = R + ) Transitions: l g,a,r l T with: g is the guard, a is the label, r is the set of clocks to be reset to 0 Semantics: States: (l, v) where v is a valuation for the clocks Action transition: { (l, v) 0 g,a,r a (l, v l ) iff l A, v = g, v = v[r 0] Delay transition: t R +, (l, v) t (l, v + t) iff 0 t t, v + t = Inv(l)

50 Model checking for TA A defines an infinite timed transition system. Decision procedures are based on the region graph technique (Alur, Courcoubetis, Dill) From A and Φ, one defines an equivalence A,Φ s.t.: ( ) v A,Φ v (l, v) = Φ iff (l, v ) = Φ R + X / A,Φ is finite A region = An equivalence class of A,Φ ) X We can reduce A = Φ to (A R + / A,Φ = Φ... and use a standard model checking algorithm.

51 Model checking for TA A defines an infinite timed transition system. Decision procedures are based on the region graph technique (Alur, Courcoubetis, Dill) From A and Φ, one defines an equivalence A,Φ s.t.: ( ) v A,Φ v (l, v) = Φ iff (l, v ) = Φ R + X / A,Φ is finite A region = An equivalence class of A,Φ ) X We can reduce A = Φ to (A R + / A,Φ = Φ... and use a standard model checking algorithm.! The size of R + X / A,Φ is in O( X! M X )!

52 The region abstraction y X = {x, y} M x = 3 and M y = x

53 The region abstraction y X = {x, y} M x = 3 and M y = x compatibility between regions and constraints x k and y k

54 The region abstraction y X = {x, y} M x = 3 and M y = x compatibility between regions and constraints x k and y k compatibility between regions and time elapsing

55 The region abstraction y X = {x, y} M x = 3 and M y = x compatibility between regions and constraints x k and y k compatibility between regions and time elapsing

56 The region abstraction y X = {x, y} M x = 3 and M y = x compatibility between regions and constraints x k and y k compatibility between regions and time elapsing

57 The region abstraction y X = {x,y} M x = 3 and M y = region defined by I x =]1;2[, I y =]0;1[ {x} < {y} x compatibility between regions and constraints x k and y k compatibility between regions and time elapsing

58 The region abstraction y X = {x,y} M x = 3 and M y = x region defined by I x =]1;2[, I y =]0;1[ {x} < {y} successor by delay transition: I x =]1;2[, I y = [2;2] compatibility between regions and constraints x k and y k compatibility between regions and time elapsing

59 An example [AD 90 s] y x

60 Complexity of timed verification Reachability in TA is PSPACE-C (Alur & Dill) Reachability in TA with three clocks is PSPACE-C (Courcoubetis & Yannakakis) Reachability in TA with constants in {0, 1} is PSPACE-C (Courcoubetis & Yannakakis) Model checking Timed CTL over TA is PSPACE-C (Alur, Courcoubetis & Dill) Model checking AF Timed µ-calculus over TA is EXPTIME-C (Aceto & Laroussinie) ( But tools (Kronos, UppAal) exist and have been applied ) successfully for verifying industrial case studies.

61 One/two clock timed automata [LMS2004] 1 Model checking TCTL on 1C-TA is PSPACE-complete. 2 Reachability in 1C-TA is NLOGSPACE-complete. 3 Model checking TCTL, on 1C-TA is P-complete. 1 Reachability in 2C-TA is NP-hard. 2 Model checking CTL on 2C-TA is PSPACE-complete.

62 Outline 1 Timed specification languages 2 Timed models Discrete-time models Timed automata 3 Conclusion

63 Conclusion discrete time dense time DTG 0/1 DTG [LMS06] 1C-TA 2C-TA TA [ACD93] [LST03] jump cont. [LMS04] [LMS04] [CY92] Reachability NLOGSPACE-C NP-hard PSPACE-C TCTL, P-complete PSPACE-C TCTL P-compl. p 2 PSPACE-complete TCTL c PSPACE-complete

64 Conclusion No efficient algorithm for TCTL+clocks. As soon as there are two clocks, a complexity blow-up occurs for any verification problem. For TCTL, model-checking may be efficient only for the simple DTGs with durations in {0, 1}. For TCTL,, it is possible to have efficient model-checking algorithm for any kind of DTG and 1C-TA. The previous result can be extended to probabilistic timed systems (Probabilistic DTG and 1-clock Probabilistic TA).