Planning with sharable resource constraints
IJCAI'95 Proceedings of the 14th international joint conference on Artificial intelligence - Volume 2
Precedence constraint posting for cyclic scheduling problems
CPAIOR'11 Proceedings of the 8th international conference on Integration of AI and OR techniques in constraint programming for combinatorial optimization problems
| Hi-index | 0.00 |
In this paper we develop and analyse CSP-based procedures for solving scheduling problems with metric temporal constraints (e.g. jobs' deadlines or separation constraints between couple of consecutive activities in a job) and multiple capacitated resources, referred to formally as the Multiple Capacitated Metric Scheduling Problem (MCM-SP). This work follows two different solution approaches used to identify resource capacity conflicts. (1) Profile-based approaches (e.g. [1]) - They consist of characterizing a resource demand as a function of time, identifying periods of overallocation in this demand profile, and incrementally performing "leveling actions" to (hopefully) ensure that resource usage peaks fall below the total capacity of the resource. (2) Clique-based approaches [3, 2] - Given a current schedule, this approach builds up a conflicts graph whose nodes are activities and whose edges represent overlapping resource capacity requests of the connected activities. Fully connected subgraphs (cliques) are identified and if the number of nodes in the clique is greater than resource capacity a conflict is detected. Clique-based approaches perform more global analysis and offer greater accuracy in conflict detection in comparison with local pairwise profile-based analysis, but at a potentially much higher computational cost. In a previous paper [1], several profile-based solution procedures were developed and evaluated (in particular the ESTA algorithm), some of them descending from an approach proposed in the planning literature. Here we complete our analysis with a set of experiments which evaluate cost/performance tradeoffs on problems of increasing scale, varying both the tightness of temporal constraints and the size of resource capacity levels. In problem space regions which are highly constrained, clique-based approaches produces more and better quality solutions, even if they have a much higher computational cost than profile-based approaches, and the differential increases as resource capacity levels increase. In this less constrained circumstance, both methods perform comparably wrt solution quality, suggesting the advantage of the profile-based approaches over the clique-based ones.