Improved approximation algorithms for scheduling parallel jobs on identical clusters

Marin Bougeret (1) , Pierre-Francois Dutot (2, 3) , Denis Trystram (3, 2, 4) , Klaus Jansen (5) , Christina Robenek (5)

1 MAORE - Methods, Algorithms for Operations REsearch
2 UGA [2016-2019] - Université Grenoble Alpes [2016-2019]
3 MOAIS - PrograMming and scheduling design fOr Applications in Interactive Simulation
4 IUF - Institut universitaire de France
5 Department of Computer Science [Kiel]

The Multiple Cluster Scheduling Problem corresponds to minimize the maximum completion time (makespan) of a set of n parallel rigid (and non-preemptive) jobs submitted to N identical clusters. It cannot be approximated with a ratio better than 2 (unless P = NP). We present in this paper the methodology that encompasses several existing results [1, 2]. We detail first how to apply it for obtaining a 5 -approximation. Then, we use it to provide a new 7 -approximation 23 running in O(log (nhmax)N(n + log(n))), where hmax is the processing time of the longest job. Finally, we apply it to a restriction of the problem to jobs of limited size, leading to a 2-approximation which is the best possible ratio since the restriction remains 2-inapproximable.

Mots clés: Scheduling; Parallel job; Strip packing; Approximation algorithm

Domaines: Calcul parallèle, distribué et partagé [cs.DC]

Marin Bougeret, Pierre-Francois Dutot, Denis Trystram, Klaus Jansen, Christina Robenek. Improved approximation algorithms for scheduling parallel jobs on identical clusters. Theoretical Computer Science, 2015, 600, pp.70-85. ⟨10.1016/j.tcs.2015.07.003⟩. ⟨hal-01230293⟩