docPart I - Chess - Center for Hybrid and Embedded Software Systems
download 
Report Transcription
Part I - Chess - Center for Hybrid and Embedded Software Systems
Predictable Timing of Cyber-Physical Systems Future Research Challenges DREAMS Seminar, EECS, UC Berkeley January 17, 2012 David Broman [email protected] EECS Department UC Berkeley, USA Department of Computer and Information Science Linköping University, Sweden 2 Agenda [email protected] Part I Semantic gap regarding time Part II Bridging the gap – the PRETIL project Part I ! Semantic gap regarding time ! Part II! Reducing the gap – the PRETIL project! Part III Utilizing a bridged gap – virtual optimization of CPS Part III Utilizing a bridged gap – virtual optimization of CPS! 3 [email protected] Part I Semantic gap regarding time Part I ! Semantic gap regarding time ! Part II! Reducing the gap – the PRETIL project! Part III Utilizing a bridged gap – virtual optimization of CPS! 4 Modeling Cyber-Physical Systems [email protected] Model Equation-based model Abstraction “physical modeling” Networking Sensors System Actuators Physical system (the plant) Part I ! Semantic gap regarding time ! Part II! Reducing the gap – the PRETIL project! Embedded systems (computation) Part III Utilizing a bridged gap – virtual optimization of CPS! 5 Equation-Based Object-Oriented (EOO) Languages Domain-Specific Language (DSL) [email protected] Models and Objects • Primarily domain: Modeling of physical systems • Multiple physical domains: e.g., mechanical, electrical, hydraulic Equation-Based Object-Oriented (EOO) Part II! Reducing the gap – the PRETIL project! Part I ! Semantic gap regarding time ! • Object in e.g., Java, C++: object = data + methods • Objects in EOO languages: object = data + equations Part III Utilizing a bridged gap – virtual optimization of CPS! 6 Equation-Based Object-Oriented (EOO) Languages Domain-Specific Language (DSL) • Primarily domain: Modeling of physical systems connections! objects (components)! Models and Objects Equation-Based ports! Object-Oriented (EOO) • Multiple physical domains: e.g., mechanical, electrical, hydraulic EOO model (textual) Part I ! Semantic gap regarding time ! [email protected] Part II! Reducing the gap – the PRETIL project! • Object in e.g., Java, C++: object = data + methods • Objects in EOO languages: object = data + equations EOO model (graphical) Part III Utilizing a bridged gap – virtual optimization of CPS! Equation-Based Object-Oriented (EOO) Languages Domain-Specific Language (DSL) 7 [email protected] Models and Objects • Primarily domain: Modeling of physical systems • Multiple physical domains: e.g., mechanical, electrical, hydraulic • Object in e.g., Java, C++: object = data + methods Equation-Based Object-Oriented (EOO) Acausality • Objects in EOO languages: object = data + equations • At the equation-level u=R*i • At the object connection level Part II! Reducing the gap – the PRETIL project! Part I ! Semantic gap regarding time ! Part III Utilizing a bridged gap – virtual optimization of CPS! Equation-Based Object-Oriented (EOO) Languages Domain-Specific Language (DSL) [email protected] Direction not determined at modeling time! • Primarily domain: Modeling of physical systems • Multiple physical domains: e.g., mechanical, electrical, hydraulic Models and Objects Equation-Based acausal (non-causal) Object-Oriented (EOO) Acausality causal Part I ! Semantic gap regarding time ! 8 Part II! Reducing the gap – the PRETIL project! • Object in e.g., Java, C++: object = data + methods Variables ! • Objects in EOO languages: ! Potential! object = data + equations ! Flow! Physical topology ! • At the equation-level u = Ris*lost! i • At the object connection level Part III Utilizing a bridged gap – virtual optimization of CPS! Equation-Based Object-Oriented (EOO) Languages Domain-Specific Language (DSL) • Primarily domain: Modeling of physical systems • Multiple physical domains: e.g., mechanical, electrical, hydraulic • Object in e.g., Java, C++: object = data + methods Equation-Based acausal (non-causal) Object-Oriented (EOO) causal • Objects in EOO languages: object = data + equations • At the equation-level u=R*i • At the object connection level Part II! Reducing the gap – the PRETIL project! Part III Utilizing a bridged gap – virtual optimization of CPS! Equation-Based Object-Oriented (EOO) Languages Domain-Specific Language (DSL) • Primarily domain: Modeling of physical systems • Multiple physical domains: e.g., mechanical, electrical, hydraulic • Modelica • VHDL-AMS • gPROMS • MKL • (SPICE) Part I ! Semantic gap regarding time ! [email protected] Models and Objects Acausality Part I ! Semantic gap regarding time ! 9 10 [email protected] Models and Objects Equation-Based Object-Oriented (EOO) Acausality • Object in e.g., Java, C++: object = data + methods • Objects in EOO languages: object = data + equations • At the equation-level u=R*i • At the object connection level Part II! Reducing the gap – the PRETIL project! Part III Utilizing a bridged gap – virtual optimization of CPS! 11 Modeling Cyber-Physical Systems [email protected] Platform 1 Actuator Physical Interface Physical Plant 2 Network Physical Interface Platform 2 Sensor Computation 1 Delay 1 Computation 4 Computation 2 Sensor Platform 3 Physical Interface Delay 2 Computation 3 Physical Plant Plant 12 Physical Model Physical Interface Actuator Different models of computation Equation-based model Abstraction “physical modeling” C-code Networking Sensors System Actuators Physical system (the plant) Part I ! Semantic gap regarding time ! Part II! Reducing the gap – the PRETIL project! Embedded systems (computation) Part III Utilizing a bridged gap – virtual optimization of CPS! 12 Modeling the Systems for Computing and [email protected] Networking Ptolemy II Heterogenous modeling environment supporting many different models of computation (MoC). For example, synchronous dataflow (SDF), discrete-event (DE), process networks (PN), etc. PTIDES Currently implementation in Ptolemy. Modeling of event-based realtime distributed systems. Based on DE semantics. Synchronous reactive languages For example, Lustre, Signal and Esterel Next versions of Modelica New semantics for synchronouse discrete semantics (for improved code generation). Simulink And all other languages/environments not listed here! Part I ! Semantic gap regarding time ! Part II! Reducing the gap – the PRETIL project! Part III Utilizing a bridged gap – virtual optimization of CPS! 13 Simulation the CPS [email protected] Platform 1 Actuator Physical Interface Physical Plant 2 Network Physical Interface Platform 2 Sensor Computation 1 FMI Delay 1 Computation 4 Computation 2 Sensor Platform 3 Physical Interface Delay 2 Computation 3 Physical Interface Physical Plant Plant 12 Physical Model Actuator Different models of computation Equation-based model C-code Networking Sensors System Actuators Embedded systems (computation) Physical system (the plant) Part I ! Semantic gap regarding time ! Part II! Reducing the gap – the PRETIL project! Part III Utilizing a bridged gap – virtual optimization of CPS! 14 Simulation the CPS [email protected] Platform 1 Actuator Physical Interface Physical Plant 2 Network Physical Interface Platform 2 Sensor Computation 1 Delay 1 Computation 4 Computation 2 Software-in-the-loop (SIL) simulation Different models of computation Sensor Platform 3 Physical Interface Delay 2 Computation 3 Physical Plant Plant 12 Physical Model Equation-based model Hardware-in-the-loop (HIL) simulation Physical Interface Actuator Code generation Physical system available? Networking Sensors System C-code Actuators Physical system (the plant) Part I ! Semantic gap regarding time ! Part II! Reducing the gap – the PRETIL project! Embedded systems (computation) Part III Utilizing a bridged gap – virtual optimization of CPS! 15 Simulation the CPS [email protected] Platform 1 Actuator Physical Interface Physical Plant 2 Network Physical Interface Platform 2 Sensor Computation 1 Delay 1 Predictable Timing of Cyber-Physical Systems Software-in-the-loop Computation 4 Computation 2 Sensor Platform 3 Physical Interface Delay 2 (SIL) simulation meaning that the continuous-time timing behavior for Different models of computation Equation-based model SIL simulation ҮHIL simulation ҮReal-time system execution Code Hardware-in-the-loop generation Note that predictability is a continuum. The “cyber” can be (HIL) simulation C-code made deterministic, but the physics cannot. Computation 3 Physical Plant Plant 12 Physical Model Physical Interface Actuator Physical system available? Networking Sensors System Actuators Embedded systems (computation) Physical system (the plant) Part II! Reducing the gap – the PRETIL project! Part I ! Semantic gap regarding time ! Part III Utilizing a bridged gap – virtual optimization of CPS! 16 Physical Model problems Model and Timing Problems Incorrect Parameters Control Delay Cyber timing problems Modelica / MKL Incorrect System of Equations Validation of models From sampling to actuation Jitter [email protected] Ptolemy II / PTIDES Large model libraries. Mature tools. Computation Precision-timed machines Clock sync, IEEE 1588 Hard to predict. E.g., loss data packets. Related to robustness. (Wittenmark et al., 1996) Part I ! Semantic gap regarding time ! Precision-timed - Scratchpad memory - Timing instructions - Thread-interleaved piplines Part II! Reducing the gap – the PRETIL project! Automatic allocation of scratchpads? Semantic gap regarding time Variation of start times (e.g., clock accuracy, architecture) Transient Errors Timingconstraints of tasks WCET of tasks Communication Bounded delays How to ensure that compilation is semantically correct regarding time? PRET Part III Utilizing a bridged gap – virtual optimization of CPS! 17 [email protected] Part II Bridging the gap – the PRETIL project Part II! Reducing the gap – the PRETIL project! Part I ! Semantic gap regarding time ! Part III Utilizing a bridged gap – virtual optimization of CPS! 18 Precision-Timed Intermediate Language (PRETIL) [email protected] High-level requirements Modelica / MKL Ptolemy II / PTIDES Make code generation from source language to PRETIL simple (e.g., via suitable API) Other MoC and tools Support multiple modeling (source) languages Expose language constructs for (physical) execution time PRETIL Hide (abstract away) architecture dependent details (e.g., scratchpad) PRETIL compiler PRET PRET PRET Part I ! Semantic gap regarding time ! Part II! Reducing the gap – the PRETIL project! Formal semantics – reason about correctness of execution time General purpose CPU PRET PRET Enable comparison of platforms Part III Utilizing a bridged gap – virtual optimization of CPS! 19 Execution time – a correctness factor [email protected] Worst-case execution time (WCET) Best-case execution time (BCET) Estimated upper bound of WCET Challenge to make it tight Part I ! Semantic gap regarding time ! Part II! Reducing the gap – the PRETIL project! Part III Utilizing a bridged gap – virtual optimization of CPS! 20 Sketch - primitives for handling time (pseudo-code, part of research to be performed) [email protected] F(x1,…,xn) is a function in the language with n parameters. ! Dynamic usage of execution time Static Usage of execution time Execute with padding without guarantees Propagating WCET info up the tool chain: - For meta-programming (static scheduling) - For tool support (e.g., show WCET for specific actors in Ptolemy) execute f(3,2) during 10ms else …! Use WCET/BCET info dynamically in the model/program. if WCET(f) > 10ms then … else …! Propagate time constraint downwards constraint WCET(f) < 10ms! Execute with padding (exact time) WCET of parameterized functions in runtime using parametric WCET analysis (Lisper, 2003) execute f(3,2) during 10ms! Part I ! Semantic gap regarding time ! Part II! Reducing the gap – the PRETIL project! if WCET(f(x3 = v)) > 10ms then … else …! Part III Utilizing a bridged gap – virtual optimization of CPS! 21 Proposed Infrastructure Overview [email protected] Part I: Modeling language front end Part II: PRETIL front end Part III: PRETIL backend Part IV: Runtime environment Part I ! Semantic gap regarding time ! Part II! Reducing the gap – the PRETIL project! Part III Utilizing a bridged gap – virtual optimization of CPS! 22 Part I – Modeling language front end [email protected] Research challenge 1: To design (or extend) an intermediate language that hides architecture details and exposes language constructs for programming with (physical) execution time. Part I ! Semantic gap regarding time ! Part II! Reducing the gap – the PRETIL project! Part III Utilizing a bridged gap – virtual optimization of CPS! 23 Proposed Infrastructure Overview [email protected] Part I: Modeling language front end Part II: PRETIL front end Part III: PRETIL backend Part IV: Runtime environment Part I ! Semantic gap regarding time ! Part II! Reducing the gap – the PRETIL project! Part III Utilizing a bridged gap – virtual optimization of CPS! 24 Part II – PRETIL Front end [email protected] Research challenge 2: To statically guarantee that timing constraints defined for high-level models hold during run-time. Formally verified compilers (Leroy, 2009) Part I ! Semantic gap regarding time ! Translation Validation Infrastructure (Necula, 2000) Part II! Reducing the gap – the PRETIL project! Part III Utilizing a bridged gap – virtual optimization of CPS! 25 Proposed Infrastructure Overview [email protected] Part I: Modeling language front end Part II: PRETIL front end Part III: PRETIL backend Part IV: Runtime environment Part I ! Semantic gap regarding time ! Part II! Reducing the gap – the PRETIL project! Part III Utilizing a bridged gap – virtual optimization of CPS! 26 Part III – PRETIL Back end [email protected] Research challenge 3: To optimize allocation of bounded memory resources so that both memory constraints and timing constraints hold simultaneously. Part I ! Semantic gap regarding time ! Part II! Reducing the gap – the PRETIL project! Part III Utilizing a bridged gap – virtual optimization of CPS! 27 Proposed Infrastructure Overview [email protected] Part I: Modeling language front end Part II: PRETIL front end Part III: PRETIL backend Part IV: Runtime environment Part I ! Semantic gap regarding time ! Part II! Reducing the gap – the PRETIL project! Part III Utilizing a bridged gap – virtual optimization of CPS! 28 Part IV – Runtime environment [email protected] Research challenge 4: To guarantee safe execution concerning timing of a deployed binary of machine code, without trusting the correctness of the compiler, e.g., by executing a lightweight safety proof before executing the binary. Proof-carrying code (Necula, 1997) Part I ! Semantic gap regarding time ! Part II! Reducing the gap – the PRETIL project! Part III Utilizing a bridged gap – virtual optimization of CPS! 29 [email protected] Part III Utilizing a bridged gap – virtual optimization of CPS Part I ! Semantic gap regarding time ! Part II! Reducing the gap – the PRETIL project! Part III Utilizing a bridged gap – virtual optimization of CPS! 30 Simulation with Predictable Timing [email protected] Platform 1 Actuator Physical Interface Physical Plant 2 Network Physical Interface Platform 2 Sensor Computation 1 Delay 1 Computation 4 Computation 2 Software-in-the-loop (SIL) simulation Different models of computation Sensor Platform 3 Physical Interface Delay 2 Computation 3 Physical Plant Plant 12 Physical Model Equation-based model Physical Interface Actuator The PRETIL project aims at adding one piece of the puzzle to getting predictable timing of CPS Networking Sensors System Actuators Physical system (the plant) Part I ! Semantic gap regarding time ! Part II! Reducing the gap – the PRETIL project! Embedded systems (computation) Part III Utilizing a bridged gap – virtual optimization of CPS! 31 Optimization with Predictable Timing [email protected] Platform 1 Actuator Physical Interface Physical Plant 2 Network Physical Interface Platform 2 Sensor Computation 1 Delay 1 Computation 4 Computation 2 Software-in-the-loop (SIL) simulation Different models of computation Sensor Platform 3 Physical Interface Delay 2 Computation 3 Physical Plant Plant 12 Physical Equation-based model Physical Interface Actuator Design optimization problems • Parameter optimization of physical objects (e.g., thickness of shafts) • Architecture parameters, e.g., minimize clock frequency to lower energy consumptions. Predictable timing with correct timing constraints are essential to performing the optimization on a global CPS model Hard problems. One approach is to combine CPS simulation with local search heuristics (e.g., tabu search or simulated annealing). Part I ! Semantic gap regarding time ! Part II! Reducing the gap – the PRETIL project! Part III Utilizing a bridged gap – virtual optimization of CPS! 32 Conclusions and Summary [email protected] Modelica / MKL Ptolemy II / PTIDES New project in the Ptolemy group (starting Jan 2012). Overall challenge To establish a new formal foundation of timing predictability for the semantics of correct translation/ compilation from high-level CPS modeling languages down to machine code for PRET machines. Thank you for listening! Part I ! Semantic gap regarding time ! Part II! Reducing the gap – the PRETIL project! Semantic gap PRETIL regarding time PRETIL compiler PRET Part III Utilizing a bridged gap – virtual optimization of CPS!
