Cooperative parallel decomposition guided VNS for solving Weighted CSP

Abdelkader Ouali (1) , Samir Loudni (2) , Lakhdar Loukil (1) , Patrice Boizumault (2) , Yahia Lebbah (1)

1 LITIO - Laboratoire d'Informatique et Technologies de l'Information d'Oran
2 Equipe CODAG - Laboratoire GREYC - UMR6072

Tree decomposition introduced by Robertson and Seymour aims to decompose a problem into clusters constituting an acyclic graph. Recently, Fontaine et al. [8] introduced DGVNS (Decomposition Guided VNS) that uses the graph of clusters provided by a tree decomposition to manage the exploration of large neighborhoods. However, for large scale problems, the performance of DGVNS may decrease significantly due to the large number of clusters to be considered sequentially. To overcome this shortcoming we propose CPDGVNS (Cooperative Parallel DGVNS) in which the clusters are explored in parallel through an asynchronous master-slave architecture. Experiments performed on real life instances show the appropriateness and the efficiency of our approach.

Mots clés: Tree decomposition; Weighted CSP; Parallelization; Meta-heuristics; Variable Neighborhood Search (VNS); Master-Slave architecture

Abdelkader Ouali, Samir Loudni, Lakhdar Loukil, Patrice Boizumault, Yahia Lebbah. Cooperative parallel decomposition guided VNS for solving Weighted CSP. 9th Int. Workshop on Hybrid MetaHeuristics (HM 14), Jun 2014, hamburg, Germany. pp 100-114, ⟨10.1007/978-3-319-07644-7_8⟩. ⟨hal-01024659⟩