Title: University of Colorado at Denver 2006 Math Clinic Dynamic Scheduling of Prioritized Tasks
1University of Colorado at Denver2006 Math
ClinicDynamic Scheduling of Prioritized Tasks
2Agenda
- Raytheon initiatives and interests
- Satellite Scheduling
- Dynamic Tasking
- Math Clinic problem statement review and
expectations - QA
3Raytheon Initiatives and Interests
4More Than 30 Years of Space Experience
Mission Management
Command and Control
Ground Terminals
Data Processing
Network Management
Complete ground capabilities for space systems
5Scheduling Research Interests
- Scheduling Topic
- Near Real-time Dynamic Re-Tasking Systems
- Dynamic scheduling of prioritized tasks
- Potential Scheduling Approaches
- Meta Heuristics
- Parallel/Distributed Scheduling
- Continuous Optimization
- Scheduling via Data Clustering
- Scheduling via Predictive Analysis
- Automated Optimizer Tuning (feedback loops)
- Clinic should take a fresh approach, and leverage
the 2005 Math Clinic work where possible.
6ExampleMission Scheduling of a Notional
Commercial Remote Sensing Satellite System
7Satellites
Remote Sensing
Scientific Discovery
Intelligence
Communications
8Introduction
- Satellite mission scheduling is the process of
assigning resources to tasks, over time. - Resource sensor, receiver, thruster, gyro,
antenna, processor - Task collection, downlink, orbit adjustment,
data processing function - Payload scheduling attempts to optimally sequence
a collection of related activities on a payload
resource (e.g. sensor), to maximize the efficient
use of this expensive, oversubscribed resource
Tasks to be Scheduled
The Schedule
Resource
Scheduler/Optimizer
TN
T0
Time
9Modeling
- Basics
- Tasks
- Activities to be performed (image collection,
orbit adjust) - Occupy discrete time on a schedule (start time
duration) - Resources
- Objects that perform activities (sensors,
thrusters) - Constraints
- Restrict when, where and how tasks are scheduled
- Tasking (input parameters preferred times,
minimum quality) - Visibility (access to target pointing/attitude
time) - Availability (resource availability or capacity)
- Temporal (time order relationships between
tasks) - Physical (power restrictions, thermal exposure)
10Tasks
- Task
- Id
- Collection Interval
- Earliest latest collection times
- Priority (numeric)
- Target
- Cartesian coordinate system
- Duration (sec)
- Strategy
- Collection Strategies
- Simple
- A single collection for a single target
- Monitor
- Periodic (revisit) collections over a defined
interval of time for a single target - Initial task model
- Iterate over the course of the semester, if
necessary
- Complexity
- 1000?3000 potential collections per area of
interest - A mix of standing tasks and on-demand requests
- 1?N on-demand requests, from M users, that arrive
according to some probabilistic distribution
during the course of an executing schedule
11Visibility Constraint to Target
- Shape size of orbit determine when and how
often a point on the earth is visible from the
satellite - Satellite agility also plays a role in target
access times - Role (smaller)
- Role Pitch
- Role, Pitch Yaw (larger)
- Collection activity can only be scheduled during
a Window of Opportunity
12Slewing
- Slewing is the act of changing a satellites
attitude (pointing), typically performed when
transitioning from the collection of one target
to another. - Generally slewing introduces idle time for the
sensor - Consider a shortest path problem
- Visiting targets that are close together ?
shorter slews ? less idle time - When scheduling, slewing is only one
consideration. One must balance a shortest
(bore-sight) path with other criteria, such as
task priority, collection strategy and collection
quality.
13Quality Constraints
- There can be a relationship between geometries
and collection quality, which in turn affects
collection duration. - Angle between satellite and target
- Angle between sun and target
14Dynamic Tasking
15Traditional TCPED Cycle
16New TCPED Cycle with Dynamic Tasking
17Dynamic Tasking Description
- Interpretation is a new decision function
executed in near real-time - Can be performed anywhere (processing facility,
in the field, etc) - Can be performed by human or machine
- Results in a decision versus analysis and
reporting products (exploitation) - Causes near real-time redirection of assets
(dynamic re-tasking) - Implications to Payload Scheduling
- Larger input data size due to many more sources
- Non-deterministic arrival of tasks (on-demand
tasking) - Much shorter response times (rapid rescheduling)
18Example of Near Real-time Dynamic Tasking Systems
Notional Surveillance Attack on enemy tanks
(timeline does not reflect required, designed or
actual capabilities)
Dynamic Re-tasking
Satellite Finds Tanks 000030
Satellite Tracks Tanks 000045
Data Passed to commander 000200
Commander Tasks Aircraft 000300
Initial Satellite Tasking 000000
Satellite Re-Tracks Tanks 000100
Enemy Tanks Destroyed 001300
Kill Chain 13 minutes from first observation to
bombs on target
19Example of Near Real-time Dynamic Tasking Systems
20Problem Description Review and Expectations
21Problem Description
- Schedule tasks for a single asset (could use
commercial remote sensing satellite as an example
test case) - New task requests can arrive at any time
(non-deterministic) - Most task requests are due in near real-time
- All tasks have a fixed duration
- Assume that there is access for all tasks over
the scheduling interval (1 hour access interval) - Assume that there is a transition cost between
any two tasks - The time to transition from one task to another
is a function of the distance between the task - Assume only one task is performed at a time
22Problem Description Details
- Tasking
- Id (alpha-numeric)
- Activity Interval
- Earliest Start from 5 sec - 30 min out, from task
arrival - Latest End from 10 sec - 1 hour out, from task
arrival - Earliest Start ? Preferred Start ? Preferred End
? Latest End - To schedule outside the Preferred interval is a
soft constraint violation with a penalty that
increases linearly the farther you are from the
Preferred interval. Scheduling outside the
Earliest/Latest interval is a hard constraint
that cannot be violated. - Priority (0 - 100, where 100 is the lowest
priority 0 is the highest) - location (non-uniformly distributed - clusters)
- Duration (fixed at 5 seconds for all tasks)
- Strategy (simple or monitor)
- Activity Strategies
- A mix of more Simple and fewer Monitor
- Scheduling Goals
- Activities will not miss their due dates
- Higher priority tasks take precedence over lower
priority tasks - A densely packed schedule
- Build an existing schedule and then reschedule
based on dynamic re-task requests
23Problem Description Details
- Scheduling Objectives (samples)
- Maximize the sum of the priorities AND
- Minimize transition costs AND
- Maximize -utilization of schedule
- Priorities are not cumulative i.e. one priority
10 task is more important than ten priority 100
tasks. - Characterizations of scheduling performance
(samples) - Average runtime vs the number of tasks scheduled
- Average wait time from task input to task
execution - -satisfaction of tasks (i.e. scheduled vs
considered) - Clinic should identify appropriate
characterizations. Focus on performance from a
high volume, dynamic user input perspective.
24Expectations Deliverables
- Literature search (annotated bibliography)
- Chosen algorithm
- Justification
- Description
- Performance/Characterization results
- Prototype (Matlab, Java, C/C)
- Point of Contact
- Mick Stahlberg
- mjstahlberg_at_raytheon.com
- 720.858.4204
25Questions Answers