Model-Based Computing for Design and Control of Reconfigurable Systems Markus Fromherz, Daniel Bobrow, Johan de Kleer

1
Model-Based Computing for Design and Control of
Reconfigurable SystemsMarkus Fromherz, Daniel
Bobrow, Johan de Kleer

• Presented By Darlene Banta

2
What is a Reconfigurable System?

• Complex electro-mechanical products (high end
  printers and photocopiers)
• Designed as families with reusable modules put
  together in different manufacturable
  configurations
• Modules controlled locally by software that must
  take into account the entire configuration

3
The Problem

• How to Build a system control framework for
  systems that can be configured out of different
  modules
• To decide at design time whether a proposed
  module is a worthwhile addition to the system

4
The SolutionQualitative Constraint Based Models

• Primary Task is CONTROL
• Approach to the design, control and evaluation of
  complex systems
• Software engineers become model builders
• Planning and constraint satisfaction replace
  program execution
• The control software in the machine explicitly
  constructs a plan and monitors its execution in
  real time
• Hierarchical Control Software Architecture to
  mirror the architecture of the machine itself

5
Solution Goals

• Increase productivity of software developers
• Improved communication among different subsystem
  engineers (mechanics, electronics, software, etc)
• Ensure consistency across different engineering
  tasks (design, control, testing, diagnosis)
• Enable automatic modular configuration of the
  resulting systems of the controller

6
System Controller

• Breaks systems functions down into module
  functions and coordinates the modules to produce
  the desired documents
• Determine operations that will complete the task
  successfully
• Optimize machine productivity
• Generate and commit to schedules incrementally

7
1. Modeling

• Set of connected components, each contains
• Structural Description
• Behavioral Description
• Declarative modeling approach to use models for
  other tasks
• Need to provide the necessary information to
  enable the control task
• Planning identify component capabilities in
  sequence
• Scheduling feasible timings for capabilities in
  a plan
• Derives constraints from the physical structures
  of devices

8
(Module Modeling)

• Specify module components and their connections
• Define itineraries the mappings from module
  commands to component commands
• Integrates component control to high-level
  commands
• Control Module receives a command and sends
  required component commands to its components

9
2. Planning

• Forward simulation with discrete events and event
  propagation
• Creates a sequence of modules that should be
  visited in order
• Possible to cache the results for various plans,
  thus creating an efficient part of the process

10
3. Scheduling

• Solves time constraints in order to find a
  feasible schedule while maximizing productivity
• Generate a schedule incrementally
• Scheduling Algorithms and Architectures
• Amount of look-ahead
• Timing of when to commit to parts of the
  incremental schedule

11
4. Modeling LanguageComponent Description
Language (CDL)

• Declarative specification of input/output
  constraints
• Provides
• Behavioral statements
• Specification of structural elements
• Based on C Syntax
• Translate language into concurrent constraint
  programming (CCP)
• Simulation
• Partial Evaluation
• Abduction
• General Reasoning

12
How It Works Theoretically

• Machine adjusts to its own machine model
• Modules pass up their module models to the system
  controller
• At the system controller module models composed
  to a machine model
• At run time, given the machine model and
  specifications, the system controller plans the
  module operations that need to be executed then
  schedules these operations

13
Configuration Analysis

• Individual users subject the system to workload
  distributions that vary from the average
• Given a machine model with a set of design
  variables whose values are unknown, the design
  and analysis task consists of determining
  consistent values for these variables to optimize
• Performance
• Cost of resulting design
• Model based techniques develop declarative,
  multi-use models of machines with certain
  parameters open and only constrained by ranges
• Using these models, design solutions can be
  analyzed considering Workload Distribution
• Exact
• Qualitative

14
Exact and Qualitative Workload Distributions

• Exact
• Model based scheduler determines and compares the
  productivity of a set of configurations
• Qualitative
• Compute the expected deviations of different
  configurations from optimal productivity
• Determine a convex hull that shows which designs
  are best for which workload distributions

15
Implementation

• Solver Alternates Between
• Adding Constraints
• Searching for solutions
• Committing to partial solutions
• Solver manages relation between timing variables
  and real time
• Full optimization with minimal commitment
• Optimizer returns a solution for the next sheet
  only but guarantees it is part of a currently
  optimal solution for all known sheets

16
Solver Design

• Low-level constraint operations should be
  incremental and distributed over time to minimize
  their efforts
• Allows trade-offs between memory and processor
  usage
• Scheduling algorithm should be able to make use
  of its application model to help the solver
  manage its resources efficiently

17
Real Time

• Does not impose deadlines on jobs there is
  always a feasible schedule
• Always finds a solution in polynomial time for
  typical machines and jobs

18
Benefits

• Better Basis for
• Reusability
• Compositionality
• Represent capability execution as discrete events
  with predictable durations and transport times
• Enables qualitative and quantitative reasoning at
  design and run time

19
Benefits

• Models describe the local behaviors of components
• States constraints and transformations on
• Parts moving through the components
• Constraints on the timing of resource allocation
• Combination of symbolic, qualitative and
  numerical constraints and connecting component
  models to describe the entire machines
• Reason about the behavior of the composite
  configuration
• Enables a variety of applications