ADASS the Planning and Scheduling Perspective by ewghwehws


									                                                                                                         Q uic kT im e™ an d a
                                                                                             T IFF ( Un co m pr e ss e d) d ec o mp r es s or
                                                                                                 a re n ee d ed to s e e th is p ictu r e.

                                       ADASS the Planning and Scheduling
                                         - How planning and scheduling fits in at ADASS
                                         - ADASS planning and scheduling posters and
                                         - Invited talk I gave on scheduling research I am
                                         doing with Mark Johnston

            Q uic kT im e™ an d a
T IFF ( Un co m pr e ss e d) d ec o mp r es s or
    a re n ee d ed to s e e th is p ictu r e.
           Planning and Scheduling
•   ADASS = Astronomical Data Analysis and Software Systems
     – Main focus of the conference is on post observation data handling
       and processing
     – How does planning and scheduling fit in?
•   Many of the people who attend ADASS operate telescopes
     – Realize that planning and scheduling impacts their lives
        • Lots of people were interested in planning and scheduling
     – ADASS = Astronomical (Data Analysis and Software Systems)
•   ADASS planning and scheduling papers give the “
•   ” as opposed to the “state of the art”.
      – Papers describe what missions are doing as opposed to the latest
        developments in planning and scheduling.
      – For the state of the art in planning and scheduling try the
        International Conference on Automated Planning and Scheduling
        (ICAPS) or IWPSS
          Posters and Presentations
•   The GBT Dynamic Scheduling System: A New Scheduling Paradigm
     – Ground based system that dynamically schedules observers a few
       days in advance based on long term constraints and predicted
     – Papers discussed how this worked with users and technical
       approaches (e.g. knapsack problem)
•   Planning and Executing Airborne Astronomy Missions on SOFIA
     – Telescope mounted on the side of an airplane
     – Have worked on scheduling techniques for individual missions
     – Interested in long range planning techniques
•   Mission Science Operations at the Southwest Research Institute in
    Boulder, Colorado
•   Planning and scheduling within the WSO-UV observatory
     – How to adapt an existing plan
     – Mixed mode automated planning and realtime operations
            Invited Talk Summary
• Research in multi-objective
  multi-participant scheduling.                   multi-objective optimizer   • science and

• Goal: to increase science                                                     resource optimized

• Provide tools which allow       • objectives
                                  • constraints
  operators to make tradeoffs                          • alternatives
                                  • compromises        • tradeoffs
  between competing               • decisions          • optimized
  objectives                                             schedules

• Work done with Mark
  Johnston at JPL                                                       missions
                                      multiple participants
       enable participants to work
Tools do not support collaboration wrong
  Cartoon Example of what can go
between participants
togetherscheduling process.
  in the

                                Scheduling input given in
                                isolation. Schedules do notcreate
                                Participants collaborate to meet
                                users needs meet their needs
                                schedules that

 Our goal is to provide decision support tools that enable multiple
 participants to optimize schedules in a collaborative manner.
       Multi-Objective Scheduling-
•   Effective scheduling of missions requires the ability to make trade-offs
    between competing objectives:
     – Time on target, minimizing use of consumables, minimizing the use
        of critical mechanisms, preferring the higher priority science
•   Objectives are often competing in that improving one objective means
    making another objective worse.
•   Objectives have different constituents lobbying for them
     – e.g. Mission science community versus Engineering
•   The traditional approach is to combine the weighted average of separate
     – (Obj1 * wt1 + Obj2 * W2 …. + Objn * Wn) / n
     – Combining objectives loses information and pre-determines trade-offs
       between objectives.
       Multi-Objective - Solutions
• Multi-Objective Scheduling:
   – Explicitly maintain and exploit multiple objectives during
     scheduling - Don’t combine objectives
   – Algorithms build up approximate Pareto optimal frontier
      • i.e. “non-dominated” solutions, such that no other candidate
        is better, considering all objectives.
      • Utilizing evolutionary algorithms (e.g. GDE3)

               QuickTime™ and a                     QuickTime™ and a
               PNG decompr essor                    PNG decompr essor
        ar e neede d to see this picture.    ar e neede d to see this picture.
    • The Pareto frontier gives participants an optimal trade-off space
    • Still need to agree on a particular candidate schedule
    • Multi-participant tools will provide distributed decision support
         – Mixed-initiative planning – support the end user in making trade offs
             • Automate when possible but leave final control with the user
         – Graphical internet-based tools that support multiple participants
         – Challenges include: human factors, non-simultaneous users,
           domain-specific scheduling GUIs

  Proposed model:
threaded news/mail
      reader +
  schedule viewer
        JWST Scheduling Results

- Pareto Optimal surfaces for each pair of objectives
- Evaluated alternative search evolutionary approaches
•   Saturn orbiter + Titan lander
     – launched 1997
     – arrived at Saturn 2004
•   Science instruments include
    6 for optical and microwave
    remote sensing, and 6 for fields/particles/waves investigations
•   Spectacular scientific success
     – 260 scientists from 17 countries participating
     – science objectives coordinated by 6 science discipline-oriented
        teams: Rings, Atmospheres, Titan, Icy Satellites,
        Magnetosphere, and Cross-Discipline (everything else)
•   ~1 Gigabyte per day science data returned
•   Prime mission completed; currently in first 2 year extension of prime
    mission: a second 2 year extension is expected
Multi-Objective Cassini Science Planning: Example
                                                   Future Work
                                                                                 Q uic kT im e™ an d a
                                                                     T IFF ( Un co m pr e ss e d) d ec o mp r es s or
            Q uic kT im e™ an d a                                        a re n ee d ed to s e e th is p ictu r e.
T IFF ( Un co m pr e ss e d) d ec o mp r es s or
    a re n ee d ed to s e e th is p ictu r e.

               • Develop multi-participant capabilities
                  – Threaded email model
                  – DSN scheduling as an application
               • New capabilities in framework
                  – Parallel evaluation of evolutionary algorithms
               • Apply framework to other applications
                  – Planning HST phase 1 observations?
                  – JWST long range planning?

To top