Safer PI ISIS May 00 Seattle

Slide1: 

Model-Integrated Approach to Adaptive Embedded Systems Akos Ledeczi Institute for Software Integrated Systems Vanderbilt University

Outline: 

Outline Key benefits of MIC Static (design-time) adaptivity Dynamic (run-time) adaptivity Conclusions

Benefits of MIC: 

Benefits of MIC Level of abstraction: source code vs. high-level system models Constraints: implicit vs. explicit constraints Scope: point solution vs. design space

System Models: 

System Models Domain-specific language Primarily graphical Multiple-aspects Hierarchical decomposition Model: representation of all the information that is necessary to automatically synthesize the application

Constraints: 

Constraints Domain-specific: model-integrity constraints Application-specific: compositional resource performance cost etc.

Scope: Design-Space: 

Scope: Design-Space Explicit alternatives andgt;andgt;andgt; finite design space Parametric, generative models: architectural parameters algorithmic description andgt;andgt;andgt; infinite design space

Design-Space Exploration: 

Design-Space Exploration Step 1: Search-space pruning symbolic constraint satisfaction (e.g. OBDDs) interactive and iterative (selective constraint relaxation) input: 10n (n andgt;andgt; 1); output: 1-10 configurations Step 2: Simulation Step 3: Optimization Step 4: Automatic system synthesis

Automatic System Synthesis: 

Automatic System Synthesis Hardware configuration (e.g. VHDL) Source code (C, C++, Java, SQL etc.) Configuration information: real-time schedules message routing maps assignment etc.

Slide9: 

Target Application - Formal constraints MIC Approach System Models System Synthesis

Slide10: 

S1 S3 S2 e1[S21]/ / /../ /../ (D1.time - D2.time) andlt; 2 P3 (mode=(S1 or S2))implies(P1=P1i) Pr1 Pr3 C1 if(k = 2 and n andlt; 4) generate(n) else generate(k) Functional models Resource models Behavior models Explicit alternatives Performance constraint Compositional constraint Hierarchical Parallel FSM G n k Generator Parameters Explicit alternatives Design Space Modeling

Slide11: 

Design Space Pruning

Slide12: 

Modeling objects are lost at generation step: Fixed configuration or mode-based run-time adaptivity only No incremental changes No feedback to model Static (Design-Time) Adaptivity Models Synthesis Config / source files Application

Slide13: 

Embedded synthesizer Dynamic (Run-Time) Adaptivity Models Synthesis Application Embedded models replace (augment) configuration / source files: self-describing system incremental change capability system state is communicated in terms of model-based information

Adaptive Behavior: 

Adaptive Behavior Monitoring Analysis, redesign, verification Reconfiguration

Monitoring: 

Monitor variables reflect system state in the model What to monitor? Kernel/OS status Application state Events, exceptions Monitoring Embedded synthesizer

Analysis, Redesign, Verification: 

The evaluator is a system process Varying complexity: table-driven model-defined logic procedural etc. Generator scripts: Controlled by their input parameters in the model Cannot legally be bypassed to change model Reliably handle transitions, sequencing and incremental changes Constraint manager Analysis, Redesign, Verification Embedded synthesizer R/T Executive process 2 process 3 process 4 Evaluator Constraint Manager Generator scripts Evaluator

Reconfiguration: 

Reconfiguration Embedded synthesizer The embedded synthesizer is a system process Relatively simple translation

Conclusions: 

Conclusions High-level system models and explicit constraints: postpone implementation decisions as late as possible Design space modeling and automatic system synthesis: amortize initial investment over the lifetime of the application Embedded models and synthesis: self-describing system