A Hierarchical Approach for Resource Management in Heterogeneous Systems

Heterogeneous architectures are emerging as a dominant trend for HPC, mainly thanks to their high performance-per-watt ratio. Dealing with heterogeneity and task-based applications requires to consider different aspects at both infrastructures level and single node in order to meet power, thermal and performance requirements. Thus, in order to provide an effective and fine-grained management of the available resources, as well as balancing the load by dispatching applications among the different computing nodes, we proposed a hierarchical approach in which different resource managers, running in the nodes, collaborate to reach a multi-objectives optimization.