Model-driven performance prediction of distributed real-time embedded defence systems Katrina Falkner Nickolas Falkner James Hill Dan Fraser Marianne Rieckmann Vanea Chiprianov Claudia Szabo Gavin Puddy Adrian Johnston Andrew Wallis
May 30, 2015
Model-driven performance prediction of distributed real-time embedded defence systems
Katrina Falkner Nickolas Falkner James Hill Dan Fraser Marianne Rieckmann Vanea Chiprianov Claudia Szabo Gavin Puddy Adrian Johnston Andrew Wallis
Agenda
• Model-driven engineering and System execution modelling for defence systems
• The architecture of the performance prediction system
• Early validation on an Unmanned Air Vehicle (UAV)
• Conclusion and perspectives
University of Adelaide 2
Model-driven engineering and System execution modelling for defence systems
• Requirements of DRE defence systems
– Long life-cycles
– Change in development philosophies
– Modular design
– Reuse
– Greater concern for non-functional
• Space, weight, power
University of Adelaide 3
Model-driven engineering and System execution modelling for defence systems
• Performance prediction while(!perfModel.satistify(userPerfGoal)){
perfModel<-improvedPerfModel;
}
• Model-driven engineering
– Model
– Execute
• System execution modelling (SEM)
– Performance specificity
– Hardware testbeds
University of Adelaide 4
The architecture of the performance prediction system
University of Adelaide 5
Modelling
• Modelling the System under study (SUS) – the SEM
– Systemic structure
– Functional behaviour
– Workload
– Deployment
• Modelling Scenarios
– Simulate realistic interactions
– Analyse performance of SUS
– Scenario Domain Specific Language (DSL)
University of Adelaide 6
Executing
• Executing the System execution model (SEM)
– Application: SEM + scenarios
– Middleware: Data Distribution Service DDS
– Operating system
– Hardware
• Executing Scenarios
– Platform specific information
– Code generation of distributed units
– Deployment
University of Adelaide 7
Defence needs
Evaluating and predicting
• Collect execution traces
• Aggregate metrics
• Evaluate if(perfModel.meet(
perfConstraints))
• Visualize
University of Adelaide 8
Early validation on an Unmanned Air Vehicle • Scenario:
=> change in bandwidth
=> change in CPU workload
University of Adelaide 9
UAV in the air
UAV going underwater
Early validation on an Unmanned Air Vehicle
University of Adelaide 10
Systemic structural model of the SUS
Behavioural and workload models of the SUS
Early validation on an Unmanned Air Vehicle
• Evaluating utilization:
u =𝑠𝑒𝑟𝑣𝑖𝑐𝑒 𝑡𝑖𝑚𝑒
𝑟𝑢𝑛𝑡𝑖𝑚𝑒
uAIR=4.15%
uSUB=59.6%
for workload=150 msec
University of Adelaide 11
Execution traces of the SEM
Conclusion and perspectives
• Model-driven performance prediction system
– Integration of realistic data sources
– Visualization of the causes of performance issues
– Understanding of models and relationships
• Perspectives
– Graphical Scenario DSL
– Performance DSL
– Multi-modelling DSL
University of Adelaide 12