for the High-performance Modeling

Size: px

Start display at page:

Download "for the High-performance Modeling"

Robert Morton
7 years ago
Views:

1 1 Unifying Framework and Tool for the High-performance Modeling and Verification of Hybrid Concurrent Systems Kazunori Ueda Dept. of Computer Science and Engineering, Waseda University and National Institute for Informatics November 2009

2 Research Groups and Their Relationship 2 Verification and Dependability Unifying Languages (LMNtal) High-Performance Verification Parallel Processing Hybrid Systems (HydLa) with NII Three interrelated groups Two cross-cutting concerns

3 Unifying Programming Language 3 Project LMNtal (pronounce elemental ) A concurrency model + language + system Started in 2002, now working on the 4G impl. >100, LOC involving many people over the years Focused on verification (state-space space search, LTL model checking) since 2007 Provides an IDE with visualization Ready to use; very low entry barrier Papers: [Ueda, RTA 08] [Ueda et al., ICTAC 09] [Ueda, TCS 410 (2009)]

4 LMNtal: What and Why 4 Rule-based concurrent language for expressing & rewriting connectivity and hierarchy Substrate model of various calculi (λ, π π, ambient, etc.) Computation is manipulation of diagrams Links express 1-to-1 connectivity Membranes express hierarchy, locality and first-class multisets Allows programming with sets and graphs and programming by self-organization Well-defined notion of atomic actions

5 LMNtal allows us to represent computation in terms of hierarchical graph rewriting 5 X links X0 A nodes i o many-to-1 comm. n n L0 S1 S b n right L1 n L2 n Y operation buffer header A hub formed by membrane Y0 s m m X X Y g m Z Y channel formed by membrane send protected by membrane receive m asynchronous π-calculus Y m m X Y unprotected closed unary i o function i o + m c c m + X cyclic structures map function

6 Related work: Models and languages with multisets and symmetric join 6 (Coloured) Petri Nets Production Systems and RETE match Graph transformation formalisms CCS, CSP Concurrent logic/constraint programming Linda Linear Logic languages Interaction Nets Chemical Abstract Machines Gamma model Maude Constraint Handling Rules Mobile ambients P-system, membrane computing Amorphous computing Bigraphs

7 Models and languages with membranes + hierarchies 7 (Coloured) Petri Nets Production Systems and RETE match Graph transformation formalisms * * : some versions CCS, CSP feature hierarchies Concurrent logic/constraint programming Linda * Linear Logic languages Interaction Nets Chemical Abstract Machines Gamma model Maude Constraint Handling Rules Seal calculus l Mobile ambients Kell calculus P-system, membrane computing Brane calculi Amorphous computing κ Bigraphs

8 Demo: Water Jug Problem 8 Typical AI search problem E.g. use a 300ml jug and a 500ml jug to get 400ml of water Operations:? Empty a jug 300ml 500ml 400ml Fill up a jug with tap water Move water until it s emptied ed Move water until the other is filled Fill up the Move to the

9 Towards Model Checking in LMNtal 9 Computer-aided verification deals with systems for which LMNtal allows concise description: state transition systems (automata) multiset rewriting systems concurrent systems LMNtal is at the same time a full-fledged fledged programming language. No gap between modeling and programming languages (cf. SPIN, nusmv,...) As a fine-grained concurrent language, supporting verification is highly desirable Why not build an integrated development and verification environment?

10 MC in LMNtal: strengths and challenges 10 The LMNtal IDE supports the understanding of models with and without errors, not just bug catching workbench for designing and analyzing models complementary to fast, black-box checkers Challenge: implementing state management graph isomorphism: a notoriously hard problem in general

11 Experiences with the LMNtal model checker 11 Applications so far real-time scheduler security / data transfer protocol AI search checking of the fine-grained, graph encoding of the untyped lambda calculus [Ueda, RTA 08] etc. Multiset rewriting allows concise encoding of problems (e.g., n-queens) and state-space (symmetry) reduction (e.g., philosophers) Visualization turned out to be very useful for understanding systems

12 High-Performance Verification 12 High-performance parallel computing applied to computer-aided verification Parallel SAT (propositional satisfiability) solver First step: c-sat (cluster-based scalable solver) [Ohmura and Ueda, SAT 09] Engine for verification and AI applications Model checking, AI planning, constraint satisfaction,... Parallel model checkers Algorithms (task partitioning, i etc.) Systems (multi-core LMNtal underway) Few research groups worldwide

13 c-sat: Parallel SAT Solver for Clusters 13 Question : How scalable is distributed-memory parallel SAT based on modern solvers? SAT search is notoriously irregular. Goal : comprehensive evaluation of c-sat on Beowulf clusters with PEs (cores) Result: >23x speedup p (geometric mean) with 31 PEs for 189 hard instances from SAT Races / Competition hundreds/thousands of machine hours >31x / >19x for SAT / UNSAT instances

14 Scalability (31 PEs, hard instances) SAT 1024 UNSAT Spee edup rat tio linear speedup MiniSat runtime (s) lower bounds

15 Hybrid Systems 15 Involves both continuous and discrete changes billiard, air-conditioning, crossing gate,... Programming language aspects rather unexplored Designing and implementing HydLa, a constraintbased language for modeling hybrid systems Key challenges are the semantics and validated computation under uncertainties Interdisciplinary field, ranging over logic, constraints, numerical computation and programming languages Relates also to biology and control engineering

16 HydLa: Overview and Features 16 Declarative ( procedural) employs popular math and logical notations Constraint-based dff differential equations and other constraints define functions over time able to handle partial information (numerical errors, uncertainties, etc.) Features constraint hierarchies It s difficult to describe systems so that the constraints are consistent and well-defined. cf. Frame problem

17 HydLa Example: Sawtooth 17 INIT 0 f < 1. INCREASE (f = 1). t DROP (f =1 1 f = 0). INIT, (INCREASE << DROP). Describes properties at time 0 Time is implicit it (f = 1) means t 0 (f (t) = 1) Intuitive meaning of constraint hierarchy is the maximal consistent set of the constraints, subject to hierarchy and atomicity

18 Example: Crossing gate (cf. Henzinger, LICS 96) 18 TRAIN(N, Ipos, Vel) { train(n) = Ipos (train(n) ( ( ) =Vel) )} SENSOR(S, Pos, Sig) { (Sig = 0) << N S( (train(n) = Pos Sig = 1))} CONTROLLER { (raise = 0) << (raise = 1 compositionality (app ( = 1 raise^5 = 0) communication cat on (exit = 1 raise^5 = 1))} GATE { (g =0) << ( g = 90 Both symbolic and interval-based (raise = 1 g < 90 g =10) (raise implementations underway. = This 0 g is > really 0 g non-obvious = 10))} obvious due to TRAIN(1, 4000, 20), the TRAIN(2, interplay 4000, between 20)^220, SENSOR({1,2}, 1000, uncertainties app), SENSOR({1,2}, and conditionals 100, exit), CONTROLLER, GATE. [Ishii, Ueda et al., ADHS 09].

19 Computing paradigms must change Concurrency Everywhere! 20th century 21st century von Neumann architecture + sequential computation multi-core / clusters / Grid / distributed / embedded / molecular /... Turing Machines (computability) RAM model (complexity) λ-calculus (programming g languages) Floating point arithmetic (numerical analysis)

20 Grand Challenges 20 Computing paradigm as a reliable and accessible infrastructure to prepare for more challenging and sophisticated apps Unify programming computation on symbolic and modeling and verification verfcaton and numerical processing

21 Research Collaboration 21 Possible key principles: constraints (Univ. Nantes, LINA) Other possible common backgrounds: concurrency, nondeterminism hybrid systems verified software Our (relative) strength: language and system implementation parallelization Need of collaboration: reaching out to neighboring i disciplines i s inside and outside CS (e.g., control and biology) exploring unforeseen applications

22 Thanks for your attention! 22

Model Checking based Software Verification

Model Checking based Software Verification 18.5-2006 Keijo Heljanko Keijo.Heljanko@tkk.fi Department of Computer Science and Engineering Helsinki University of Technology http://www.tcs.tkk.fi/~kepa/ 1/24