Modelling and analysing real-time systems with MAST: Overview and future extensions

Transcription

1 Modelling and analysing real-time systems with MAST: Overview and future extensions By: Michael González Harbour Universidad de Cantabria, Spain FACULTAD DE CIENCIAS 7/Feb/11 1

2 1. Introduction: Objectives of MAST Develop a model for describing the timing behavior of event-driven distributed real-time systems - composable software modules - separation of architecture, platform, and software modules Open model that may evolve to include new characteristics or points of view of the system Develop a set of tools for analyzing the timing behavior of the application FACULTAD DE CIENCIAS 7/Feb/11 2

3 MAST Environment Model Building Tools Component Profile Ada Profile MARTE Profile Data handling Tools Results viewer Graphical editor XML converter RT Model Data Model description Results description Trace log Tool launcher Tools Simulator Sensitivity Prio. assignment Blocking time analysis Schedulability analysis FACULTAD DE CIENCIAS 7/Feb/11 3

4 2. Real-Time Model: Overview Software Modules Processing Resource Shared Resources Scheduler Platform Operation Scheduling Server Real-time situation Activity Handler Timing Requirement Scheduling Parameters Schedulable Entities Reference FACULTAD DE CIENCIAS 7/Feb/11 4

5 2. Real-Time Model: Overview Software Modules Schedulable resource Processing Resource Shared Resources Scheduler Platform Operation Scheduling Server Mutual exclusion resource Step Real-time situation Activity Handler Timing Requirement Scheduling Parameters Schedulable Entities Reference FACULTAD DE CIENCIAS 7/Feb/11 5

6 Real-Time Situation Transaction (End-to-end Flow) External (Step) (Step) (Fork) Activity Activity Multicast Internal (Workload ) Handler Handler Handler Timing Requirement Handlers Transaction... FACULTAD DE CIENCIAS 7/Feb/11 6

7 3. Elements of the MAST model: Example RMT: Teleoperated Robot Command Message Ethernet Network Status Message Remote Station Local Controller 1sec GUI 50ms Trajectory Planner Command Manager 5ms Servo Control Reporter Data Sender FACULTAD DE CIENCIAS 7/Feb/11 7

8 Processing resources, schedulers, drivers, and timers overheads packet driver ethernet packet-based network packet driver remote_station primary scheduler ethernet primary scheduler local_controller primary scheduler remote_station local_controller regular processor timer timer regular processor FACULTAD DE CIENCIAS 7/Feb/11 8

9 Scheduling servers (schedulable resources) priority trajectory planner data sender reporter gui command manager servo control ethernet driver1 message stream ethernet driver2 remote_station primary scheduler ethernet primary scheduler local_controller primary scheduler FACULTAD DE CIENCIAS 7/Feb/11 9

10 Shared (mutual exclusion) resources Name Shared_Resource Immediate_Ceiling_Resource Priority_Inheritance_Resource Ceiling Preassigned SRP_Resource Preemption_Level Preassigned FACULTAD DE CIENCIAS 7/Feb/11 10

11 Operations Overridden_- Sched_Parameters 0..1 Overridden_Sched_Parameters Name Operation Composite_Operations_List 1..* Composite_Operation Simple_Operation Execution_Times 0..* 0..* To_Lock To_Unlock Shared_Resource Message_Transmission Transmission_Times Enclosing_Operation Execution_Times FACULTAD DE CIENCIAS 7/Feb/11 11

12 GUI operation in the example includes set_command simple operation gui enclosing operation includes read_status simple operation gui() { while (true) {... set_command();... read_status();... wait_period(); } } command uses uses status immediate ceiling resource immediate ceiling resource FACULTAD DE CIENCIAS 7/Feb/11 12

13 Transactions (end-to-end flows): Distributed transaction in the example e1 event pattern timing reqs. timed activity Op: tr_planning Srv: tr_planning activity e2 activity (step) Op: command Srv: msg_stream activity e3 activity Op: Comm_mgmt Srv: Comm_mgmt activity e4 o1 Op: reporter Srv: reporter e6 Op: status Srv: msg_stream e5 Op: data_sender Srv: data_sender FACULTAD DE CIENCIAS 7/Feb/11 13

14 External (workload) s External_ Periodic_ Period Max_Jitter Phase Aperiodic_ Avg_Interarrival Distribution Singular_ Phase Sporadic_ Min_Interarrival Unbounded_ Bursty_ Bound_Interval Max_Arrivals FACULTAD DE CIENCIAS 7/Feb/11 14

15 Handlers Activity (Step) / Rate Divisor / Delay / Offset Concentrator (Merge) Barrier (Join) Delivery Server (Branch) Multicast (Fork) FACULTAD DE CIENCIAS 7/Feb/11 15

16 Timing requirements Timing_Requirement Composite_Timing_Requirement Simple_Timing_Requirement 1..* Requirements_List Deadline_Timing_Requirement Deadline 1 Referenced_ Max_Output_Jitter_Req Max_Output_Jitter FACULTAD DE CIENCIAS 7/Feb/11 16

17 4. Hierarchical schedulers in MAST Scheduling Parameters Scheduling Parameters Scheduling Parameters Scheduling Scheduling Server Scheduling Server Server Scheduling Parameters Scheduling Parameters Scheduling Parameters Scheduling Scheduling Server Scheduling Server Server Scheduling Policy Secondary Scheduler Scheduling Policy Primary Scheduler Processing Resource Scheduling Server Reference Scheduling Parameters FACULTAD DE CIENCIAS 7/Feb/11 17

18 5. The MAST Tools Automatic calculation of blocking times Schedulability analysis - fixed priorities - EDF - heterogeneous systems - hierarchical scheduling Automatic assignment of scheduling parameters - priorities - deadlines Sensitivity analysis Discrete-event simulator FACULTAD DE CIENCIAS 7/Feb/11 18

19 Fixed Priority Response-Time Analysis Technique Single- Processor Multi- Processor Simple Transact. Linear Transact. Classic Rate Monotonic Varying Priorities Holistic Offset Based Unoptimized Offset Based Multiple Multiple T. FACULTAD DE CIENCIAS 7/Feb/11 19

20 EDF Response Time Analysis Tools Technique Single- Processor Multi- Processor Simple Transact. Single Processor EDF_Within_Priorities Holistic local EDF Holistic global EDF Offset-based local Offset-based global Linear Transact. Multiple T. FACULTAD DE CIENCIAS 7/Feb/11 20

21 Scheduling parameters assignment Technique Fixed priorities EDF Heterogeneous Single- Processor Monoprocessor Multi- Processor HOPA Simulated Annealing HOSDA HOSPA FACULTAD DE CIENCIAS 7/Feb/11 21

22 6. Integration into the design process Architectural real-time models Scheduling policies WCET evaluation Translation Design Detailed Design Mechanism Design Coding Architectural Design Unit Testing Concurrency patterns Synchronization patterns Generation of detailed realtime models Object analysis Integration and Test Testing Validation System Engineering Party Requirements Analysis Analysis Mapping real-time properties to subsystems High-level real-time analysis Schedulability analysis Priority Assignment Sensitivity analysis Identification of realtime situations: -Transactions - Timing requirements - Work loads FACULTAD DE CIENCIAS 7/Feb/11 22

23 7. Future Work in MAST-1 Implement missing tools: Multiple-event analysis Full support for EDF Speed up the response time analysis Enhance error reporting FACULTAD DE CIENCIAS 7/Feb/11 23

24 Future Work in MAST-2 Align names with MARTE Partitioned scheduling support for ARINC 653 systems with hierarchical scheduling - fixed priorities on top of timed partitions support for TTP networks Network switches support for AFDX deterministic ethernet Resource reservations virtual resources as a new primary scheduler FACULTAD DE CIENCIAS 7/Feb/11 24

25 Future Work in MAST-2 (cont d) Enhance modelling capabilities support for thread locking from a transaction - enable modelling synchronous RPC enhanced modelling of timers - allow multiple timers FACULTAD DE CIENCIAS 7/Feb/11 25

26 8. Conclusions MAST defines a model for describing real-time systems distributed and multiprocessor complex synchronization and event-driven schemes composable software modules independence of architecture, platform and modules MAST provides an open set of tools hard and soft real-time analysis automatic blocking times, priority assignment, sensitivity analysis... XML specification language allows easy integration with other tools (i.e., UML tools) FACULTAD DE CIENCIAS 7/Feb/11 26

27 Conclusions (cont d) MAST is free software MAST is evolving and will soon cover aspects such as: alignment with MARTE partitioned scheduling network switches additional modelling capabilities FACULTAD DE CIENCIAS 7/Feb/11 27

28 URL FACULTAD DE CIENCIAS 7/Feb/11 28