Search

International audience; In this paper, we introduce a generalization of the vector bin packing problem, where the bins have variable sizes. This generalization can be used to model virtual machine placement problems and in particular to build feasible solutions for the machine reassignment problem. We propose several families of greedy heuristics for this problem and show that they are flexible and can be adapted to handle additional constraints. We present structural properties of the machine reassignment problem, that allow us to decompose it into smaller subproblems and adapt our heuristics to them. We evaluate our heuristics on academic benchmarks of the vector bin packing problem, a randomly generated vector bin packing problem with heterogeneous bins benchmark as well as Google’s realistic instances of the machine reassignment problem.

International audience; We propose a novel job scheduling approach for homogeneous cluster computing platforms. Its key feature is the use of virtual machine technology to share fractional node resources in a precise and controlled manner. Other VM-based scheduling approaches have focused primarily on technical issues or extensions to existing batch scheduling systems, while we take a more aggressive approach and seek to find heuristics that maximize an objective metric correlated with job performance. We derive absolute performance bounds and develop algorithms for the online, non-clairvoyant version of our scheduling problem. We further evaluate these algorithms in simulation against both synthetic and real-world HPC workloads and compare our algorithms to standard batch scheduling approaches. We find that our approach improves over batch scheduling by orders of magnitude in terms of job stretch, while leading to comparable or better resource utilization. Our results demonstrate that virtualization technology coupled with lightweight online scheduling strategies can afford dramatic improvements in performance for executing HPC workloads.

We propose a novel job scheduling approach for homogeneous cluster computing platforms. Its key feature is the use of virtual machine technology to share fractional node resources in a precise and controlled manner. Other VM-based scheduling approaches have focused primarily on technical issues or on extensions to existing batch scheduling systems, while we take a more aggressive approach and seek to find heuristics that maximize an objective metric correlated with job performance. We derive absolute performance bounds and develop algorithms for the online, non-clairvoyant version of our scheduling problem. We further evaluate these algorithms in simulation against both synthetic and real-world HPC workloads and compare our algorithms to standard batch scheduling approaches. We find that our approach improves over batch scheduling by orders of magnitude in terms of job stretch, while leading to comparable or better resource utilization. Our results demonstrate that virtualization technology coupled with lightweight online scheduling strategies can afford dramatic improvements in performance for executing HPC workloads.; Nous proposons une nouvelle approche de l'ordonnancement des applications sur les calculateurs parallèles homogènes. Sa principale caractéristique est l'utilisation de machines virtuelles pour organiser le partage de \emph{fractions} des ressources de manière précise et contrôlée. Les approches existantes utilisant des machines virtuelles se sont principalement intéressées à des problèmes techniques ou à l'extension des systèmes de batch existants. Notre approche est beaucoup plus aggressive et nous recherchons des heuristiques qui optimisent une métrique particulière. Nous établissons des bornes de performance absolues et nous développons des algorithmes pour la version en-ligne, non clairvoyante, de notre problème d'ordonnancement. Nous évaluons ces algorithmes au moyen de simulations impliquant soit des traces synthétiques, soit de traces d'un système HPC existant. Nous comparons par ce moyen nos solutions aux algorithmes d'ordonnancement par \emph{batch} les plus classiques. Nous montrons que notre approche permet d'améliorer de plusieurs ordres de grandeur le facteur de ralentissement (\emph{stretch}) subit par les applications par rapport aux systèmes de \emph{batch}, tout en ayant une utilisation comparable ou moindre des ressources. Nos résultats montrent que l'utilisation conjointe des techniques de virtualisation et de stratégies d'ordonnancement en-ligne permet d'améliorer très significativement l'exécution des applications dans les systèmes de calcul HPC.