GitLab

For multiple reasons:

• Most of the existing workloads (SWF) contain core base resource request
• Node level is detailed enough to see core based allocation when using time sharing
• On current HPC centers, the request of resources is done by core (even if is ceiled depending on the number of core per nodes because there is no time sharing)
• Batsim workload that contains node resource requests are not portable across platforms especially for SMPI applications that are modeled at the core level

The list of feature and changes we need to implement this:

• change Batsim output to cores level (interval set of resources at core level) while keeping node information (and at the node level to be able to keep heterogeneity information).
• support new resource requests like nb_core and nb_host and remove the ambiguous nb_res.
• make the mapping work at core level (not sure of this... @mpoquet?)