© J. Christopher Beck Lecture 30: Distributed Scheduling
© J. Christopher Beck Outline Overview & Example Supply Chain Distributed Scheduling Exercise Distributed Timetabling Optimization and information engineering issues
© J. Christopher Beck Readings Gaspero, Mizzaro, & Schaerf, A MultiAgent Architecture for Distributed Course Timetabling, Sixth International Workshop on Distributed Constraint Reasoning, 2005
© J. Christopher Beck Distributed Scheduling Same as “normal” scheduling except there are multiple decision makers may control different resources and/or have different sets of jobs to execute local problems are inter-dependent Why distributed? real problem is distributed for geographical, political, or organizational reasons
© J. Christopher Beck Supply Chain Scheduling
© J. Christopher Beck Distributed Scheduling Local decisions depend on and effect external events Need to coordinate schedules with suppliers and consumers! Local decision maker has no control over others’ schedules
© J. Christopher Beck Distributed Scheduling Exercise: Resource Agent You control some units of a resource RED, YELLOW, GREEN, BLUE Your goal is to sell the resource at the highest price possible
© J. Christopher Beck Distributed Scheduling Exercise: Job Agent You have activities to perform that need resources You have $1000 to spend buying resources Your goal is to buy the resources required for all your activities for the least cost
© J. Christopher Beck Distributed Scheduling Exercise To sell/buy a resource agents must sign each others sheet noting quantity and price Once you have a verbal agreement – you must sign!
© J. Christopher Beck Distributed Scheduling Exercise Let the games begin!
© J. Christopher Beck Distributed Timetabling Problem (See Gaspero et al. paper) University classroom scheduling Each department “owns” a set of classrooms Each department has a set of courses to schedule Class size, A/V requirements, etc.
© J. Christopher Beck Distributed Timetabling Problem Doesn’t have to be distributed Departments can make their own schedules and ignore everyone else But, if they share rooms maybe everyone can have a better schedule No department is willing to give up authority for its classrooms Centralized solution is politically impossible
© J. Christopher Beck Distributed Timetabling Solution Decide on a currency one student not being able to take a course = 1 unit of currency What do they trade? roomslots: use a room for a time slot
© J. Christopher Beck Distributed Timetabling Solution Somewhat ad hoc solution each department creates its own schedule then a semi-market mechanism is used to trade roomslots to try an improve schedules You should have a general understanding of the problem solved and the techniques in this paper
© J. Christopher Beck Issues for Distributed Scheduling How much information do you share? Who has the authority to make decisions? What is the overall scheduling process? What happens when things go wrong?