Home
 
Attendees
Conference program
Registration
Location
Hotel information
Links
 
 
Previous editions
2002


    

Session MB1 - Exposé magistral II / Tutorial II

Day Monday, May 05, 2003
Room Procter & Gamble
President Wieslaw Kubiak

Presentations

14:45 Scheduling: Classical Concepts and Extensions
  Gerd Finke, Université Joseph Fourier, Grenoble, France

A short outline of the classical scheduling theory will be provided, summarizing some basic facts about the machines, task parameters and performance criteria. It is pointed out that there are three standard hypotheses: I. The set of jobs is processed on a single type of resource, namely the machines or processors. II. At any instant, at most one job is executed on a given machine. III. At any instant, a job can be processed on at most one machine. Three areas of extensions of the scheduling theory will be given where one of these rules I-III is no longer valid. In violation of I, a task may require supplementary resources, in addition to the machines. This field of resource constraint scheduling is already well established. One may distinguish resources that have to be allocated to a job during the complete production process like tools in manufacturing. Other resources are of the delivery or pick-up category (AGVs, conveyors and robot arms). Some elements of parallel machine scheduling theory with a single server will be given. Then, in more detail, so-called robotic cells will be analyzed. These cells consist of m machines arranged in a circular layout in the form of a flowshop and served by a single central robot. They were first introduced in the form of a three-machine robotic cell for the assembly of castings for truck differentials. Robotic cells produce, in general, mechanical parts that can wait at the machines as long as required before being picked up by the robot. There is a vast literature on the more restricted case, the hoist-scheduling problem, where the waiting times at the machines (the chemical baths) are limited. It is known that the robotic scheduling problem is already NP-hard for a flowshop with m machines and two or more different part types to be produced. It remains the interesting case of the m-machine robotic cell in which one wants to produce a single lot of identical parts. We shall concentrate on this case. Then the problem reduces to finding the optimal strategy for the robot moves in order to obtain the maximal throughput rate for the part. Usually, repetitive production k-cycles are considered, where a certain number k of parts are produced in a periodic fashion. The interesting conjecture by Sethi et al., the so-called one-cycle conjecture, claims that the maximum throughput rate is obtained by repeating a particular cycle, producing each time a single part. This conjecture has been stated in 1992 and we want to report its state-of-the-art for various robotic cell configurations. Let us now relax the hypothesis II, leading to the case where several tasks are processed together on a machine. A so-called batch machine is executing a group (or batch) of tasks in a single macro-operation. The processing time of such a batch is in general a function of the individual processing times of the tasks belonging to it. One of the most important functions is the maximum of the processing times, leading to a max-batch machine. Max-batch machines are frequently related in industry to burn-in operations in a furnace where the products are first grouped together and then heated. The exposure time to heat has a lower bound for each item given by its processing time. This implies that the total length of the burn-in operation is given by the maximum processing time in the batch. This relatively new area of scheduling theory will be briefly described. Then a particular class of batching problems will be considered. Based on a problem in the sheet-metal industry, the concept of a compatibility graph is introduced. The grouping is not arbitrary since only "compatible" jobs may be included in the same batch. In general, one has a compatibility graph with the tasks as the vertex set and the objective is to find the smallest number of batches (consisting of compatible jobs) in order to minimize the total execution time. This is equivalent to partitioning the graph into the smallest number of cliques. For many applications the compatibility graphs are related to interval graphs for which many combinatorial optimization problems can be solved efficiently. The relaxation of hypothesis III leads to the well-established multiprocessor scheduling theory where tasks are processed in parallel at the same time on several processors. We report some basic results and turn then to a new concept that we call multibin-object packing. It consists of packing objects into bins, but the objects require several bins to be packed into. This problem is both a relaxation of multiprocessor-task scheduling and an extension of bin-packing. Several application areas are described, in telecommunication, expertizing, file backups and drug testing. This new concept may be described as follows. Let us drop the constraint in multiprocessor scheduling that all parts of the tasks are performed simultaneously. Let us imagine we turn the Gantt chart by 90 degrees and let the tasks drop to the bottom. Replacing processors by bins and tasks by objects, one obtains a multibin packing. The solutions of multibin problems are defined by the assignment of each part of an object to a different bin. The order within a bin is of no importance. What matters is only the sum of the heights of the objects packed into the bins. Complexity results and new algorithms for these multibin packing problems will be presented.