Distributed Scheduling in Supply Chain Management Emrah Zarifoğlu

Decomposition Scheduling - very difficult, easily becomes NP-hard Scheduling - very difficult, easily becomes NP-hard Decomposition is a good method Decomposition is a good method –Decomposing the problem into smaller manageable pieces Most popular decomposition method: Most popular decomposition method: Distributed Scheduling Distributed Scheduling

Distributed Scheduling A new approach for the local decision makers make their own schedules according to their local objectives and constraints regarding the overall objectives and constraints of the whole system A new approach for the local decision makers make their own schedules according to their local objectives and constraints regarding the overall objectives and constraints of the whole system

Characteristics of the Distribued Scheduling Decision makers – “agent”s Decision makers – “agent”s –entities having the properties of autonomy to operate without human intervention, social ability to communicate with others, pro- activeness to take an initiative role and reactivity to respond to changes in the environment

Agents Local agents Local agents –Local decision making –Regular agents Global agents Global agents –Global decision making –Manager agent, mediator agent or master agent –Provide coordination –Enter global database –Initialize bidding process, select bids –Resolve conflicts among local agents

Coordination between Agents Direct coordination or coordination via global agent Direct coordination or coordination via global agent Realization of coordination – negotiation and bidding Realization of coordination – negotiation and bidding Simple bidding example: Simple bidding example: –Incoming job broadcasts bid request via global agent –Eligible local agents give bids –Global agent selects the best bid

Coordination between Agents (Cont’d) Bids may be dynamically revised (iterative bidding process) Bids may be dynamically revised (iterative bidding process) Collaboration between agents – cooperation to achieve better global objectives Collaboration between agents – cooperation to achieve better global objectives Depending on architecture and communication in the system agents may have authority on others hierarchically Depending on architecture and communication in the system agents may have authority on others hierarchically

Information Flow in the System Centralized (hierarchical) systems Centralized (hierarchical) systems –A unique master scheduler makes the schedule in the system and the local controllers implement it –Single layer centralized or multi-layer centralized Decentralized (heterarchical) systems Decentralized (heterarchical) systems

Advantages of Decentralized Systems Centralized Centralized –If global scheduler fails  system fails –A fault in a point in hierarchy affects upstream and downstream –Dictating all the duty Decentralized Decentralized –More fault tolerant  reactive scheduling –Shortened flow path  decreased risk of failure and more robust schedules –More initiative to local authorities (for human based systems)  motivation for the job

Disadvantages of Decentralized Systems Centralized Centralized –Less investment for education and technology –Not much data sharing –Unique responsible  easier fault detection –Better for global objective –Simple systems Decentralized Decentralized –Giving more initiative to local agents  more investment for education and technology –Awareness of local agents about the system  very good databases and information sharing systems (ERP, etc...) –Distributed authority  “conflict at jurisdiction” –Does not guarantee global objectivity –Hardware and software not well developed yet, congestion and deadlock in data transfer (huge traffic)

Classification of Decentralized Sytems Pure heterarchical systems Pure heterarchical systems –No hierarchy –Many times a feasible solution is enough Quasi-heterarchical systems Quasi-heterarchical systems –Between pure heterarchical and hierarchical –Single layer quasi-heterarchical and multi- layer quasi-heterarchical

Architectural Issues As number of layers increas  more tendency of a change of decision given by lower level authorities by higher levels (if global objective is dominant) As number of layers increas  more tendency of a change of decision given by lower level authorities by higher levels (if global objective is dominant) This may not be possible always due to bad flow of information This may not be possible always due to bad flow of information

Communication Issues Existence of alternative agents  bidding mechanism is more important Existence of alternative agents  bidding mechanism is more important If global objectives important  cooperation is better If global objectives important  cooperation is better Otherwise  negotiation is better Otherwise  negotiation is better

Distributed Scheduling in Supply Chain Internet-based supply webs Internet-based supply webs Bid request of customers from vendors Bid request of customers from vendors Willing vendors’ bids Willing vendors’ bids Vendor with best offer is awarded via Internet Vendor with best offer is awarded via Internet Settings Settings –One customer – multiple suppliers –Multiple customers – one supplier –Multiple customers - multiple suppliers Not much study in the literature for Internet- based supply chains Not much study in the literature for Internet- based supply chains

Algorithm SC-JI One or a few customers – multiple suppliers One or a few customers – multiple suppliers Job initiated Job initiated Customer gives orders Customer gives orders Orders are ranked by manager agent Orders are ranked by manager agent Bid request of manager agent from the vendors (local agents) Bid request of manager agent from the vendors (local agents) Bid prepared by vendors Bid prepared by vendors Best bid selected by manager agent Best bid selected by manager agent

Algorithm SC-VI Multiple customers - multiple vendors Multiple customers - multiple vendors Vendor initiated Vendor initiated Schedulable orders and vendors are grouped and linked to each other Schedulable orders and vendors are grouped and linked to each other Vendors are ranked according to their expertise Vendors are ranked according to their expertise Starting by the most expert vendor, jobs are selected by vendors Starting by the most expert vendor, jobs are selected by vendors Manager agent is a conflict resolver Manager agent is a conflict resolver After each cycle, same procedure is repeated if there are still unassigned orders After each cycle, same procedure is repeated if there are still unassigned orders Same procedure is held for all groups Same procedure is held for all groups

Critics about Algorithms Incoming job may be represented as an agent associated with a customer Incoming job may be represented as an agent associated with a customer For Algorithm SC-JI  illeteracy of manager agent about underlying bidding process is a disadvantage For Algorithm SC-JI  illeteracy of manager agent about underlying bidding process is a disadvantage For Algorithm SC-VI  instead of grouping the teams and items we may let all teams and jobs to be available to each other For Algorithm SC-VI  instead of grouping the teams and items we may let all teams and jobs to be available to each other Game theoretic approach may be very suitable for the nature of these algorithms Game theoretic approach may be very suitable for the nature of these algorithms

Conclusion Distributed Scheduling gives flexibility for problem solving Distributed Scheduling gives flexibility for problem solving Opportunities for new developments by the means of algorithms, cultural perception and education Opportunities for new developments by the means of algorithms, cultural perception and education Distributed Scheduling may be applied to a variety of systems Distributed Scheduling may be applied to a variety of systems Promising area for the researchers Promising area for the researchers