A Constraintbased Planning and Scheduling Framework for RealTime Applications

1
A Constraint-based Planningand Scheduling
Framework for Real-Time Applications

• Simon de Givry
• PLATON Center
• THALES Research Technology

2
Outline

• Thales applications
• Constraint technology strengths and limitations
• Towards a real-time planning and scheduling
  framework

3
Thales (ex Thomson-CSF)High-Tech Solutions,
Worldwide
57 Defense
43 Civil

• 50,000 employees, 30 outside France
• 7 billion of revenues, 70 international

4
Thales Applications Planning and scheduling a
critical function
Reconfigurable, shared and unspecialized resources

• The limitation is no longer the hardware but the
  task and resource management algorithm
• Planning and scheduling is a potential
  discriminator for future Thales systems

5
Weapon AllocationA complex combinatorial
planning and scheduling problem
p1
p2
p3
p4
p5
m1
DIVING MISSILE ATTACK
m2
ESJ
GROUND INSTALLATION
m3
GROUND ATTACK
m4
m5
m6
SEA-SKIMMER  MISSILE ATTACK
GROUND CLUTTER
r1
ESCORT SHIP
r2
"POP-UP" THREAT
FRIGATE
r3
ESCORT SHIP
r4
r5
SEA CLUTTER
r6
SUBMARINE
a1
a2
Preparation missile
Tracks reachability segment
Time
Shooting missile
Electric power unit
Flying missile

• Planning missiles / tracks assignments
• Scheduling resource management of launchers and
  missiles

6
Context of Thales supervision systems (e.g.
Weapon Allocation system)

• Operational context on-board and real-time
  systems
• On-line
• cyclical calls within a complex system
• Very short response time
• Memory space limit
• A valid plan/schedule at the deadline
• Strategic context Defense applications
• Long life cycle
• maintenance over 20 years
• retrofitting of systems (functional and platform
  evolutions)
• Product line
• reuse for building set of related systems
• Confidentiality and market protection

Time spaceguarantees
Robustnessto evolutions
Capitalization
In house know-how
7
Traditional Thales approachCostly at each system
evolution

• The process
• Physical constraint analysis and requirement
  elicitation
• Ad-hoc deterministic algorithm elaboration
• Algorithm tuning on a set of benchmarks
• Advantages
• Performances (quality and computation time) are
  good
• Time and space guarantees
• Drawbacks
• Costly tuning
• complete redo at each evolution of the
  specifications or the platform
• Local view for planning and scheduling
• no global model
• Risky capitalization
• no explicit model
• capitalization on the engineers

8
Constraint TechnologyModeling technology rather
than Programming technology

• Compositionality property
• Problem Model1 Model2
• Concurrency (order does not matter)
• Concurrent Constraint Multiple Models
• High level of abstraction
• Declarative language
• Global constraints, modeling complex problems
• Formal approach
• Declarative semantic of the constraints using
  First-Order Logic

9
Constraint technology the strengthsSoftware
engineering capabilities

• Reduce your development time effort
• Incrementality (no global view required,
  GroupWare)
• Extension/evolution is made simple just
  add/replace models
• Large library of ready to use constraints
• Efficient algorithms through global constraints
  (integrative technology)
• Secure your realization
• Requirement validation (formal model of the
  problem)
• Stronger validation conceptual errors instead of
  programming errors
• constraints are already validated, it's the same
  for the solver
• Capitalize on your domain
• Validated models can be directly reused
  (modularity at the model level)

10
Constraint technology the limitationsHigh
complexity and possible lack of robustness

• Constraint reasoning can be non-obvious !
• Debugging is still a difficult task
• Programming the search algorithm rather than
  specifying it
• Combinatorial problems remain difficult to solve
  !
• In limited time
• High variability in the results
• No information on the quality of the results
• No guarantee on time and space
• Does not take into account the time contract
• Does not take advantage of the evolution of
  platforms

Problem of complexity
Problem of robustness
11
Towards a real-time planning and scheduling
framework
Modeling
Efficiency
Constraint Programming
Hybrid Algorithms
EOLE
Optimization Framework
Time management / Adaptation
Anytime Algorithms Parameterized Search
Algorithms
High-level primitives Search Algorithm
LibraryTemplates of search Code generator
12
Towards a real-time planning and scheduling
framework Robustness to large combinatorial
problems

• Model refinement and redundant models
• Hybrid search algorithms
• Global search
• valid deductions
• completeness (quality assessment)
• Local search
• opportunism
• Hybrid search
• combines several local/global searches

cost
states
Ex. PesantGendreau 96
13
Towards a real-time planning and scheduling
frameworkTaking into account time and space
limits

• Space guarantee
• Only depth-first search or restricted best-first
  search
• Time guarantee
• Deadlines are guaranteed by an alarm
• Partial solution
• Exploiting time limits by controlling the search
• Dynamic time allocation to each search of the
  hybrid search
• based on a temporal policy, the total time is
  allocated to the different searches
• Local search complexity easily controlled
• Global search complexity controlled by parameters
• automatic tuning by using time estimation tools

14
Towards a real-time planning and scheduling
frameworkHigh-level abstraction for the search

• High-level primitives
• Framework
• Library of templates of search
• Automatic code generation (C, C, JAVA, )
• Vertical domain integration (Domain-oriented
  frameworks)

Requests
Quality assessment
Model
Hybridization and Control
Heuristics
Search Scheme
Search Strategy
15
The Constraint Programming approach Methodology
and First Results of Weapon Allocation

• The process
• Formal requirements of the problem, close to a
  first constraint model
• Model refinement
• Incremental validation
• The work with the Business Units
• Integrated team of domain and technical experts
• Several prototypes with Go/No Go decisions (5
  years of studies)
• Feasibility analysis
• Kinematics modeling using approximations
• First full mock-up (off-line)
• Think properties of the solution and not
  algorithms
• Size of the model 9300 var., 15500 constraints,
  350 lines of Eclair code
• Operational prototype (on-line)
• Size Model 600 var., 11500 constraints, 100 l.
  Search 300 l. Heuristics 200 l.
• Time and space guarantees
• Performances equivalent to the traditional
  approach in terms of quality and speed

16
Conclusion and research issues

• Planning and scheduling will be a discriminator
  factor in the future
• Constraint technology is becoming a key
  technology
• Software engineering properties
• modularity
• formal model
• Research issues
• extend these properties to the search
• how to control a search procedure to solve
  on-line planning and scheduling problems ?
• Operational development
• Integrated team (domain expert technological
  expert)
• Methodology is crucial
• Do not underestimate the modeling phase
• Research issue debugging is still a difficult
  task

17
References, News

• Tree Search, Local Search and Hybrid Search
• W. D. Harvey. NONSYSTEMATIC BACKTRACKING SEARCH.
  Ph. D. thesis, Stanford University, 1995
• E. Aarts, J. K. Lenstra. Local Search in
  Combinatorial Optimization. John Wiley Sons,
  1997
• G. Pesant, M. Gendreau. A Constraint Programming
  Framework for Local Search Methods. Journal of
  Heuristics, 1999
• Y. Caseau, F. Laburthe, G. Silverstein. A
  Meta-Heuristic Factory for Vehicle Routing
  Problems (Meta-Programming for Meta-Heuristics).
  In Proc. of CP-99, pages 144--158, Alexandria,
  VA, 1999
• Languages for modeling search algorithms
• P. Van Hentenryck. The OPL Optimization
  Programming Language. The MIT Press, 1998
• F. Laburthe, Y. Caseau. SaLSA A Language for
  Search Algorithms. In Proc. of CP-98, Pisa,
  Italy, 1998
• L. Perron. Search Procedures and Parallelism in
  constraint Programming. In Proc. of CP-99,
  Alexandria, VA, 1999
• J. Jourdan. Concurrent constraint multiple models
  in CLP and CC languages Toward a programming
  methodology by modelling, Informs, New Orleans,
  USA, October 1995
• EOLE French project, On-Line Optimization
  Framework for Telecom, http//www.lcr.thomson-csf.
  com/projects/www_eole (in french)
• Eclair Solver, open source version at
  http//www.lcr.thomson-csf.com/projects/openeclair
  
• Constraint Programming with time limits
• B. Cabon, S. de Givry, G. Verfaillie. Anytime
  Lower Bounds for Constraint Optimization
  Problems. In Proc. of CP-98, pp. 117-131, Pisa,
  Italy, 1998
• S. de Givry, P. Savéant, J. Jourdan, Optimisation
  combinatoire en temps limité Depth first branch
  and bound adaptatif, JFPLC'99, Lyon, France, 1999
  (in french)
• PLANET European network, On-Line Scheduling,
  http//www.lcr.thomson-csf.com/projects/planet/ols
  -tcu.html
• E. Horvitz, S. Zilberstein. Computational
  Tradeoffs Under Bounded Resources. Artificial
  Intelligence, 126(1-2), 2001
• News